This was undocumented and untested, and it is very unlikely that anyone
actually uses it.

It has also fundamentally been broken for a long time in the client
because filtering was applied _before_ sorting when building the thread
structure, so it was arbitrary (from the user's point of view) which
annotations would be kept by a `result` limit.

Codecov tells me that we are missing tests for these fields.

In the case of the "since" filter, this found an issue where `new
Date(<ann.updated value>)` was failing in PhantomJS because value
normalization lower-cased the 'T' and 'Z' chars in the ISO date string
and this caused parsing to fail. Fix this by making the `value` function
return a `Date` object which won't get normalized.

A difference between the existing test cases for the viewFilter service
and the actual inputs that are created by
`searchFilter#generateFacetedFilter` is that the actual input filter
object has empty term lists for fields that do not appear in the query.
Fix handling of this and add a test.

Babel inserts helper code before the first non-comment statement from the original
source, breaking the `@ngInject` annotation.

Rewrite the view filter service to make it easier to verify and fix a
bug in the handling of "any" queries:

The input filter object representing the parsed search query is
translated into a tree of filter primitives, where each primitive is
either a TermFilter that tests whether an annotation matches a single
(field, term) or a BinaryOpFilter that combines other filters using AND
or OR operators. The list of annotations are then matched against the
tree's root filter.

This system allows a nicer representation of "any" field queries by
expanding a query such as "foo bar" into:

  (quote:foo OR text:foo OR ...) AND (quote:bar OR text:bar OR ...)

In the process a mistake was uncovered in a test case for the "any"
filter which incorrectly allowed the previous broken implementation to
pass.

This is an initial translation of the existing viewFilter code to JS
using decaffeinate plus some manual fixups.

Use a URN derived from DC meta tags for creating equivalent doc links

“dc.relation.ispartof” is now used instead of “dc.source”
The goal of this is to identify a resource that’s part of another resource, therefore the more specific term fits better.

Use a URN derived from DublinCore meta tags for creating equivalent links across documents that provide ‘dc:source’ and ‘dc:identifier’ tags

This link is now the documentIdentifier for non-PDF cases for the document metadata. This metadata is shared with the sidebar, making this link appear in the list of links for the Search API calls

Adding support for onLayoutChange notifier configuration

- Add a doc comment to the _notifyOfLayoutChange function to clarify the
  purpose of the optional param.
- Add a comment inside the function to describe the high-level structure of the
  sidebar and how expanding/collapsing it is achieved.
- Correct a typo with "frameVisbileWidth"

Capture the client url to be used in subframes

When the client served by the browser extension injects itself into
child iframes on the current page, it needs to include the "assetRoot"
config setting in order for the client in the iframes to be loaded from
the extension rather than the default CDN URL.

Persist access and refresh tokens to localStorage.

Frame validity checks

As noted in PR feedback, the OAuth token key should be based on the
annotation service rather than just authority. The annotation service is
currently identified by (API domain, authority) but in future we'll
likely integrate the authority into the API URL for 3rd-party accounts
(eg.  "https://publisher.hypothes.is/api").

Rename for consistency with `saveToken`.

Convert search query parser to JS and add tests

Removing flags for multiple iframe support

For code in newly created JS files, arrow functions should be favored.

In the process fix an inconsistency where the "since" term was returned
as a number for hours, days, weeks, months and years but as a string for
seconds.

Convert the SearchFilter class to JS using decaffeinate and convert
methods which do not depend on any external state, which is all of them,
into plain functions.

Persist access and refresh tokens to localStorage, under the key:

  hypothesis.oauth.<authority>.token

Where `<authority>` is the domain of the annotation service and the
structure is the JSON-encoded object:

  {
    accessToken: <token>,
    refreshToken: <token>,
    expiresAt: <UNIX timestamp in ms>
  }

When an API token is requested for the first time during startup, the
existing token for the current authority is loaded from storage,
validated locally and then used if possible.

We should review whether or not to include this file in the source tree,
given that we use Yarn primarily for package management, but since it is
here, keep it up to date.