Stop propagation of the click event in the 'More'/'Less'
links which would otherwise bubble up to the
annotation card and scroll the page to that annotation.

Use a single CSS class for the annotation timestamp,
reply count and 'More'/'Less' link for truncated annotation bodies.

This gives the 'More' link a consistent style with the other links
and removes some duplicated styling.

For consistency with annotation bodies and to provide an easier hit
target on mobile, make the full width of annotation quotes clickable
to expand them.

The 'max-height' style property was not updated when the <excerpt>'s
enabled state changed from false to true which happened when switching
from annotation edit mode back to view mode.

Use ngStyle to ensure the inline styles applied to the <excerpt>'s
content element are updated as necessary in this case.

The inline controls use to cover the full height of the right-side
of excerpts, not just the visible area occupied by the 'More' and
'Less' links. This is no longer the case.

The use of 'pointer-events' here also caused the 'More' link
to be unresponsive in Microsoft Edge. There may be a difference
in behavior between Edge and FF/Chrome when nested elements
have different pointer-events values, but I haven't verified
this yet.

 * Remove the controller tests which no longer correspond
   to the implementation

 * Set default values for <excerpt> attributes needed
   by test

 * Add additional tests for toggling of the <excerpt>
   using inline controls and using properties to control
   expansion/collapse externally.

Run initialization of <excerpt> asynchronously so that any
directives inside its children have been resolved by the time
updateContentMaxHeight() is called.

In the context of the annotation list, the maxHeight
style property of the content element was set to 0 for the
annotation quotes because the 'ngRepeat' directive which
creates the DOM elements for the quotes had not run by the time
that updateContentMaxHeight() was called.

Wrapping the init logic in `scope.$evalAsync()` used to be
necessary so that any code touching the '.excerpt' element
ran after that element had been transcluded into the DOM.

However, this is no longer necessary since all the init logic
now runs inside a `scope.$watch()` callback, which is invoked
after `ngIf` has inserted the '.excerpt' element into the DOM.

 * Increase the expansion speed

 * Make the drop-shadow at the bottom of collapsed annotation bodies
   expand to the full width of the annotation card and be clickable
   to provide an easier-to-hit target for expanding collapsed
   annotations

 * Make the shadow non-clickable when hidden after expansion

If the enabled state of the annotation changes after the view
loaded, the truncation state did not update.

With the current feature flag implementation this typically happens
if the feature flags load after the session data has been received.
In future the intention is to make feature flags part of the session
data payload so this won't happen.

When embedded media in an annotation loads, this may trigger
a resize of the element's content and result in a change of whether
expand/collapse controls need to be shown, outside of Angular's
knowledge.

The DOM doesn't provide a nice way to monitor for resizing
of a specific elements (although hacks involving scroll events,
hidden <iframes> etc. exist) so here we just monitor for 'load'
events and trigger a digest in that case, as well as notifying
the containing directive whenever the collapsible state changes.

This implements the new annotation card truncation design
from https://trello.com/c/Jg0NWyF3/

The main consequence of the change is to make the annotation
cards more compact when truncation is enabled by:

 * Reducing the number of quote lines shown from 3 to 2

 * Showing the link to expand truncated quotes at the end
   of the second line of the quote, instead of on its own
   line

 * Moving the link to expand truncated annotation bodies
   outside of the annotation body and into a control
   inline with the tags list.

   This has been implemented by giving the <excerpt> control
   two modes, one that shows inline expand/collapse controls
   and a second that allows the containing directive to
   implement its own controls.

Update annotation timestamp when creating and editing annotations

When a new highlight is saved, updateViewModel() was called
immediately after making the request to the server to save
the highlight, but not after the response from the server
is received, when additional properties such as the last-updated
timestamp for the highlight, will have been set.

There were two problems here:

 1. The periodic interval for updating the timestamp was never started
    for new annotations because the initial call to updateTimestamp()
    was a no-op for annotations that had no timestamp yet.

 2. domainModel.updated was never updated after an annotation
    was saved to the server and hence the call to
    updateTimestamp() re-used the old timestamp.

(1) is fixed now by restarting the interval using the current
last-updated timestamp whenever the view model is updated.

(2) is fixed indirectly now by moving the logic to update
the timestamp into updateViewModel(), so that all updates
to it happen in one place. When an annotation is updated,
the updated domain model returned by the server is the one
passed into updateViewModel() to update it.

 * Separate out the logic for creating a decaying interval
   whose frequency depends on the age of some input timestamp,
   and updating the timestamp string on each interval.

   The decaying interval is now handled by 'time.decayingInterval'

 * Trigger a restart of the interval whenever the view model is
   updated in response to a change to the annotation

Fixes #2819
Fixes #2822

Clear selection before creating new annotation

Use Angular 1.3.x+ one-time bindings

Use one-time bindings in the annotation component's template
for expressions whose values cannot change during the lifetime
of an annotation card.

Testing using the default /stream view, without scrolling:

Initial status as of 1ceb5fcefd7fa057bdbcf1f73df57d007289531f averaged
over 30+ digest cycles:

 Digest cycle: ~30ms
 Watchers: 1686

With this commit:

 Digest cycle: ~26ms
 Watchers: 1649

Remove Moment.js

Clear any selection that exists in the sidebar before creating a new
annotation. Otherwise the new annotation with its form open for the user
to type in won't be visible because it's not part of the selection.

Add --debug arg to hypothesis-buildext

Consolidate /app and /app/features endpoints

Fix video embeds in Firefox < 45

innerText is non-standard and not supported in Firefox < 45.

Use textContent instead.

Enable video embeds to be shown fullscreen

Improve a code comment

Experience has taught us that the client needs the session
data (current user, list of groups) and set of enabled feature
flags at the same time, and also needs to invalidate them
in the same scenarios (eg. account switching).

Fetching this data via two separate requests made it more complicated
to ensure the client had a consistent view of session and feature-flag
data. To simplify things and also same a network request:

 * On the server, put the feature flag data into the session data payload.

 * On the client, use the existing central storage and cache management
   for the session data payload to manage feature flags as well.

   The features service now becomes a thin wrapper around part of
   the session state.

A downside of using the browser intl APIs to avoid the overhead
of Moment.js is that the output is implementation dependent.

Use toLocaleString()'s options argument where supported
(Firefox, Chrome, Edge, IE 11) for more control over formatting
of dates in annotation last-updated tooltips.

Safari is the only modern browser currently lacking support
for this option. For Safari, use toDateString() and toLocaleDateString()
which generates the most similar format.

The 'allowfullscreen' attribute must be set on both
the embed's <iframe> and the parent sidebar <iframe>
when the embed's main script is loaded to enable
fullscreen support.

Add media embeds feature

Client-side, replace any links to YouTube or Vimeo videos in annotation
body texts with <iframe> video embeds for those same videos.

Behind a new feature flag.

markdown.coffee: after rendering the user's markdown text to HTML and
sanitizing it, compile the HTML string into a DOM element and insert it into
the DOM using innerHTML from JavaScript, instead of using Angular's
ng-bind-html to insert the HTML string into the DOM.

This means that our CoffeeScript code in markdown.coffee has the rendered
annotation body as a DOM element, and can use the DOM API to modify it before
inserting it into the DOM.

Take advantage of this to find video links in the HTML and replace them with
embeds. This is done by a standalone media-embedder.js module that
markdown.coffee calls.

This module is designed to allow functions for generating embeds for different
kinds of URLs to be plugged in. So far functions for a few different kinds of
YouTube and Vimeo URLs are implemented, but further functions for more
media types can be added.

Moment.js is only used in a single place, for converting ISO8601 date-time strings
from the annotation's 'updated' field to user-friendly
date strings (eg. 'Wednesday 17 Dec, 18:59') for the tooltips
for the last-updated link in the annotation card.

Replacing this with functions from the built-in Date object saves
167KB from the generated bundle size (see
moment/min/moment-with-locales.min.js) and also speeds up the digest
cycle. For 20 annotation cards in the /stream view, generating the
date string used to take 6-7ms with Moment. It takes <2ms with
the built in functions in a current build of Chrome.

This still shows up as the most expensive watch expression, so
we might want to add caching as well.

jscs: Require `function () {}`

Change jscs's required style from `function()` to `function ()`.

We always require a space after the `function` keyword, and before {}
braces, but we do not allow a space between a function name and its ()
braces:

    var foo1 = function () {};  // Right.
    var foo2 = function() {};  // Wrong.
    var foo3 = function (){};  // Wrong.

    var foo4 = function foo4() {};  // Right.
    var foo5 = function foo5 () {};  // Wrong.
    var foo6 = function foo6(){};  // Wrong.

    function foo7() {}  // Right.
    function foo8 () {}  // Wrong.
    function foo9(){}  // Wrong.

This deviates from the Google style guide and jscs presets that we use.

Don't crash on non-JSON body

Enable showdown's literalMidWordUnderscores

This stops showdown from interpreting underscores in the middle of words
as <em> and <strong>, instead it treats them as literal underscores.

Example:

some text with__underscores__in middle

will be parsed as

<p>some text with__underscores__in middle</p>

Fixes #2598 (URLs with _'s in them were getting <em> tags inserted into
them, breaking the links).

Also see <https://github.com/showdownjs/showdown/issues/211>.

Small tweaks to the header for re-use in the blog