Add Java installation instructions to INSTALL.OSX.rst

Manage frontend dependencies with npm

This allows us to install all but compass locally in the project rather
than relying on global executables.

Part of #1231

Had inconsistencies with tag and tags.
Unwanted console.log was also removed

Fixes #1267

It's not reasonable to be shipping this here. We should be including
the source in the source map files or even including the source map
inline in the source map comment. The Annotator master branch does
this. For now, I'm putting the pressure on here. No source maps until
we get off this damn fork.

Fix #1265

Moved to hypothesis/firefox-addon

Fix mappings for tags field

1237 assets combination

The tag-it library is now configured to allow spaces.
However by default the window.open() function converts
the space characters in the query part of the uri to +
The streamsearch controller is now handling this conversion too.

Fixes #1254

For now. Soon we will merge this with the page search and make things
much more similar.

Make a directive for visualsearch that eases integration by taking
care of digest cycles and providing declarative binding for events.

Remove the state tracking with inSearch and will_search and replace
it by reflect the search directly in the query string.

Factor out a socket factory into a service and send the client ID only
before the first message. Configure the client ID and this service in a
new module, h.socket. Fix the base URI to end in a slash and to be the
root URL of the server.

Fix the border poking out under the active tab.

Rename it as session, since that's closer to what it's really about.

The auth drop down in the top right of the sidebar was showing a border
on the "top" entry because the first one was hidden.

Remove CSS which is redundant with visualsearch CSS; remove CSS which
serves no purpose or which is unused; place the search container inside
the inner div of the topbar; and just neaten things up.

Firefox Add-on; far easier than expected

After this bugfix, we are now storing the canonical URLs
in the URI field of the created annotations.

This update fixes the XPath creation, so that our highlights won't leak into them.

Some browsers seem to care about this. Close #1241.

This reverts commit 4eb0de8ea0de8098e2771c8a276d75434f2c99f6.

This commit removes the dtm plugin and all dtm dependent plugins from
the guest.coffee and introduces a dtm parameter inside the embed.txt
If this parameter is true (by default  it is) it loads the
DomTextMapper, TextPosition, FuzzyTextAnchors, PDF plugins.

Fixes #1233

When always on highlights and the highlighter mode were introduced in
ddf17189fbf2d428b21fd31065a1c2d2bc6ab08b the beforeAnnotationCreated
event was not fired in the guest frame. Instead, a special inject call
was added to the bridge to send the annotation and the event fired in
the iframe first. Later, when I removed this, I added the event after
the call to setupAnnotation.

The change in 3613c7d91682a438b1d243b9573cb941b26cdc43 uses the
annotationsLoaded event to notify listeners when highlights change as a
result of realizing anchors. This presents a problem because the bridge
ignores the beforeAnnotationCreated event for annotations that have
already been tagged by the bridge, but this load event causes the new
highlight to be tagged and serialized.

Move the beforeAnnotationCreated event to before the setupAnnotation
call, which is the order of things in non-highlighter mode, so that
the event is properly propagated to the sidebar iframe.

Close #1240

Close #1232

Angular won't see the cookie value when running from the extension
because the app page is served from the extension bundle and is
therefore on a different origin than the backend.

Similarly, Angular doesn't set the header when making cross-origin
requests.

Work around the issue by sending the token in responses from the
backend and setting the header ourselves.

The only way to get information about the location of the annotations
in the document is to request this information from the highlights.

This information can't be read out from the selectors, because anchoring
can place them *anywhere* in the document.

They also can't be read from the anchors, because anchors can contain
any kind of description; there is no mapping to any uniform coordinate
system.

The only way to get this information is to ask the hights for
the coordinates, after they have been rendered.

So this must be done after the second-phase of anchoring; after the
rendering of the highlights in the DOM.

So now we query this info from the highlights (upon rendering
in the host fram), and then store it back into the targets,
which are then sent back to the sidebar, so that this information
is available for sorting.

Fixes #1217.

When an annotation has not been saved yet, it does not have
a real ID. In these cases, it does not make sense to try to
send the empty ID over the RPC channel to try to display the
(yet) non-exisitent annotation card in the sidebar.

There is already some filtering for this on the other side,
but I think it makes sense to filter out invalid element
as soon as possible.