Oops

Replace the use of the websocket for scrollback with a simple
pagination scheme.

Rather than having a fixed body that scrolls, let the body take a
natural height and rewrite the infinite scroll handler to watch the
global scroll.

The infinite scroll directive is now smarter and has tests.

New version released has a my fix for height 100% html and body tags.
Restores scroll-to-annotation in PDFs.

This makes the code easier to understand because it makes explicit
the relationship to the line height of the label.

Close #2421

This change simplifies the AnnotationSync events and improves the
frame symmetry by giving the guest its own AnnotationSync setup
function, alongside its AnnotationUISync setup, that mirrors the
way its handled in the widget.

Previously, the `loadAnnotations` and `deleteAnnotations` events
were treated specially by AnnotationSync so that it didn't cause
endless loops. Instead, fully embrace the fact that the guest is
not responsible for loading or deleting annotations.

- Remove the setupAnnotation and deleteAnnotation methods, as they are
  no longer used.

- Factor out everything but the event publication of deleteAnnotation
  into a detach method that is the inverse of the anchor method.

- Subscribe to annotationsLoaded and annotationDeleted and invoke anchor
  and detach accordingly.

Like all the other calls, this should invoke its callback.

This ensures there's a digest cycle on the angular side after
the client state is synchronized and brings this into symmetry
with the other RPC calls defined by AnnotationSync.

This causes unnecessary network traffic and momentarily blanks the
ephemeral properties of the annotation set on the client.

Make setupAnnotation into a tiny compatibility wrapper around a new
method #anchor() method that returns a promise of the anchors. This
is cleaner than trying to store the anchors on the annotation itself,
which creates circular references, and it means there's no need to
strip these properties when serializes annotations across the frame
boundary.

Sort annotations by TextPositionSelector in the widget display. This
provides greater visual stability -- cards don't change positions
after anchoring -- at the cost of potentially out of order cards when
anchors have moved.

We're not displaying time in any timezones other than the browser
local one. Therefore, we don't need to parse and convert time zone
names. Moment will pull the timezone offset itself from the current
browser by using Date.prototype.getTimezoneOffset().

This eliminates 251kB (43kB gzipped) of minified code from our
application bundle. Wow!

The new choice is simpler. It does not deal with scrollable elements
below the window. On the other hand, this also makes it less liable
to choke on edge cases when scrolling the window would be just fine.

This becomes necessary as a result of the change introduced by
mozilla/pdf.js@b1c4b85d4fb7faba2b454d02d279644244bf282b

This method has existed for a long time so hopefully this won't
cause breakage for anyone on a slightly older FF.

It is included just fine by the require statements.

Nothing has relied upon this since the Store plugin was replaced
with an angular service.

This change requires us to introduce proxyquire for stubbing modules
that have a single function as their default export, such as the
scroll-into-view module.

The scroll-into-view library has no dependency on jQuery, is a proper
CommonJS module (no need for browserify-shim), handles nesting better,
and scrolls elements into the middle of the screen.

Overal, a better experience as a developer and a user.

We work on IE10+ but not IE9. Other major browsers all support it
for a while.

URI normalisation

Share a group

Group pages show a "Login to join group" link (opens in a new tab) if
you're not logged in.
If you're logged in they show a "Click to join group" button, posts to
new join() callable that adds the user to the group.
If you're already a member they show a "Share this link to invite people" link.

The groups dropdown list has new share icon links to the group's page,
for each group.

This ensures that after logging in the token getter resolves
properly.

The auth dialog waits for the background authentication check to fail,
but it shouldn't replace that promise with its own. It is perfectly
fine for the application to proceed while the user is entering
credentials.

In particular, this makes sure that a search without any authorization
can be performed even when the firstrun flag causes the login dialog
to show.

It turns out IE10 doesn't have the URL constructor and somehow I
misinterpreted the MDN docs.

The bind polyfill still isn't needed, except by PhantomJS.

It also seems silly not to just include wgxpath without npm. Save
the browserify-ing step.

https://github.com/substack/insert-module-globals/pull/46

Function.prototype.bind is supported in every browser for a long time.
It's only needed because, oddly, PhantomJS 1.8 doesn't support it.

The window.URL constructor is supported in IE10+ and every other major
browser for a while now. It's also not supported in PhantomJS 1.8,
though. The application currently doesn't work on IE < 10 anyway so
let's just remove this for now.

This should make sure that baseURI works completely even in IE.

I had started to implement the JWT Authorization Grant here but
forgot to back out those changes before I committed to just refactoring
the existing method. This was premature and also plain wrong in the
sense that the parameters were not being passed ('params' is the key
for sending GET parameters, not 'data' -- 'data' would be a separate
argument for POST).

Switch out jschannel for frame-rpc