If KaTex math rendering throws an error, load MathJax from the CDN and render
the math with MathJax.

Add KaTex javascript, CSS, and fonts to vendor folder.

Create a renderMath() function that parses sanitized markdown for math as denoted by '$$'

Modify ctrl.$render to first sanitize the markdown and then renderMath. Newly rendered
math is then trusted as HTML becuase otherwise ngSanatize removes crucial style attributes
that leave the math unreadable.

That syntax was not what I thought it was and had nasty issues when
there were no panes.

Hypothes.is 0.3.2

Conflicts:
	CHANGES.txt

@aron if those tooltip attributes are used somewhere I can't find
them so please correct me, sorry.

Fix #1490

Add an (optional) third argument to the method
`formHelpers#applyValidationErrors()` that can take an error string,
``reason``, to set on the form itself, rather than any particular
field. Remove the duplicate code in the auth directive to reset
the response errors on each field, which is handled already in the
``form-input`` directive.

Rework application bootstrap.

The embed code has for some time been the place where our application
is bootstrapped. When we did the epub.js integration, we added query
parameters to the embed.js view to control which plugins were loaded
by each instance of the application (a widget host for the reader frame
and a slimmer guest annotator for the chapter frames).

Change the injection bundle to instead include a separate bootstrap
script. This has the nice benefit of ensuring that while we still use
yepnope or a similar technique for script injection that our bootstrap
code runs in the same global context as the application dependencies
whether or not the embed script is run in a normal context or a Chrome
extension content script context. The need for an inline script created
by the embed code is no longer, since it is a separate script fetched
from the server (it's bundled in production, inside the monolithic
hypothesis-assets.js).

Instead of passing the URI of the sidebar iframe by templating the
bootstrap script, the embed script adds a link tag to the page with
``rel=sidebar`` and ``type=application/annotator+html``, which the
bootstrap script looks for. This pattern fits nicely with the existing
role of ``rel=sidebar`` in Firefox and Opera and nicely decouples the
embeding and the bootstrapping.

Further customization that used to take place in the query parameters
is now done through ``window.hypothesisRole``, which can specify an
alternate constructor, and ``window.hypothesisConfig``, which can be
a function to configure the options passed to the constructor. This
function will be useful for Single Sign On options, too, as laid out
in #1437.

The destroy script from the Chrome extension is put alongside the new
bootstrap script, as they complement one another and are not extension
specific.

Introduce zero-dependency Markdown Editor.

Covert px values in markdown-editor.scss to use em. Note: I did this for all values except min-height on .markdown-preview which for some strange reason caused a weird flicker in Firefox.

Add 'markdown-' namespace to .preview and .preview-toggle classes.

Unnested CSS in markdown-editor.scss. Everything is well namespaced now so this is no longer necessary.

Apply consistent sentence case to the title attributes of the markdown tools.

Went with Aron's suggestion of suffixing raw nodes with 'El,' and got rid of the
variable 'ourIframeSelection' as it has nothing to do with iframes and the code
could be written without the need for that variable.

input -> inputEl
textarea -> input

Code cleanup.

First pass at implementing @aron's mockups.

Fix bugs, breakout markdown into its own directive

Forgot to add markdown.coffee

Search for usernames

Remove d3 dependency

Replace the background with a grey bar using a div.

Previously it would update on every scroll or resize event fired on
either the document or body.

This improves performance while scrolling as we're no longer destroying
and recreating every tab on each call to update.

This makes the style and attribute iterators work correctly