Remove the unnecessary per-target hasDiff flag.

Extract the code for calculating the hasDiff/showDiff flags
into a separate method.

Now there are separate test cases for:
 - A single target with no diff
 - A single target with some diff
 - A single target with some diff, which is case only
 - Multiple targets, some with diffs
 - Multiple targets, some with diffs, but only in case

Furthermore, the used sample diff strings actually look like
real diff strings, for better readibility.

Also, removed some useless lines from the code.

(Thx @aron for all the suggestions.)

This restores the following corner case:

When there is a difference, but it's only about
the case of the characters, don't want to show
the visual diff by default, but we offer the
feature to the user.

I also updated the tests the reflect the specs.

onSetTool calls setVisibleHighlights so this is no longer needed.

- Fix a typo where a closure was unbound
- Always set target.hasDiff rather than having it be sometimes
  undefined and sometimes true.

Thanks @aron

Move the updateTimestamp function into the controller and call it
when render is called. If it is the first time, invoke it with a
flag that causes it to call itself, otherwise it will only update
once.

[WIP] Resurrect the removed visual diff support

Rather than watching a bunch of subproperties, the most complex of
which are the targets, deep watch the whole annotation. The targets
were being deep watched already. This just seems much easier and
clearer than debouncing the render calls, and scales to whatever
properties much change in the future. We'll re-evaluate if this
becomes a bottleneck.

The pattern of comparing the new value passed to a watcher with the old
value is idiomatic angular for avoiding the first digest. Instead, the
first digest is covered by the watch on the model reference identity.
After the initial rendering, the model will not change but the other
properties may, at which point re-rendering will occur.

Calculate the `showDiff` flag for the whole annotation,
instead of on a per-target basis.

Benefits:
 - One less variable stored inside the annotation model
 - We can more easily avoid unwanted overrides of the manually set value.

 * Move the calculations to the render funcion
 * Stop automatically `$sce.trust()`-ing the HTML code.
   (So now `$sanitize` is invoked automatically.)
   This also removes the need to use the `$sce` module,
   or to set the `trustedDIffHTML` flag, simplifying the code.
 * Use the `objectEquality` flag when `$watch()`ing the targets,
   so that we pick up changes in both the number and the contents
   of the targets

Update Annotator to 6da826c7 (typed-packaging branch)

Re-add jQuery output to the assets.yaml

... which removes the single remaining line of debug output.

Otherwise the Chrome extension doesn't get the jQuery file copied
into it's public folder. Also had to rename jQuery to include the
version string as Webassets will skip an asset entirely if the output
filename matches the content filename. This causes the embed.js loader
 to skip the loading of jQuery completely.

Terser logging

Fix #1645

Normalize for diacritics

Realtime (behind a feature flag)

Use jquery.min.js downloaded from jQuery.com

Because of the jquery.min.js change

- Using proper regexp from xregexp
- Renamed unidecode to uniFold

Thanks for @nickstenning

We have every analyzers setup for ES to do that more precisely.

Old default was term query.
Removed es options from client-side field definitions.
Cleanup old code

We judge whether to normalize or not by the variable type