Ensure that the `describe` function in `anchoring/pdf.js` fails with a
useful error if the passed range includes text outside of the PDF page's
text layer. In this case `getNodeTextLayer` will return `null` and
existing code in `describe` will throw to indicate the range starts or
ends outside of the text layer.

This will help to avoid the confusion that arose in
https://github.com/hypothesis/client/issues/1464.

This behavior is still not really useful from the end user's point of
view, but at least the error is better. Improving the actual behavior
will come later.

This has been replaced by the `TextRange.toRange` method in `text-range.js`.
The test cases are covered by the tests for `TextRange` and
`TextPosition`.

The PDF anchoring code was using several different dependencies to
to handle converting between ranges and text positions, as well as
normalizing ranges to that they start and end in text nodes.

This commit simplifies the picture by converting the PDF anchoring code
to use only the `TextRange` and `TextPosition` classes for these
purposes.

 - Change the PDF anchoring code to use the `TextRange` and
   `TextPosition` classes to handle converting a DOM range into text
   offsets in `describe`

 - Change the PDF anchoring code to use `TextRange` to handle resolving
   a (start, end) pair of character offsets within a page to a `Range`

 - Improve the type information for several related functions within the
   PDF anchoring code

Try to explain the issue with `textContent` in a better way.

`Node.textContent` has a quirk where it does not count text in comments
or processing instructions when called on an `Element` but does return
the comment / processing instruction data if called directly on a
`Comment` or `ProcessingInstruction` node.

To make conversion between `Range`s and `TextRange`s consistent in both
directions we need to always ignore text in comments/processing
instructions.

There is a difference between an empty selection and no selection at
all. In the latter case APIs such as `selection.getRangeAt(0)` will
throw an error.

Content in a PDF's annotation layer, such as link annotations, can
interfere with text selection and also cause creating annotations to
fail [1]. This commit resolves the problem by temporarily hiding annotation
layers when there is a non-empty selection. This allows the user to
select text inside link annotations (for example) while still enabling
the user to activate the link normally when there is no selection.

Fixes https://github.com/hypothesis/client/issues/1464

[1] This is because the annotation layer is outside the text layer and
    the PDF anchoring code expects the selection to be within the text
    layer.

Optimize the case where `toRange` is used to resolve (start, end)
offsets within the same element by only iterating over nodes in the
element once, instead of iterating separately to resolve the start and
end points.

Add a method that converts a text position within an element to one
where the offset is relative to some ancestor element.

This operation will be needed for replacing existing Range => (element,
start, end) conversions with a single one based on `TextRange` and
`TextPosition`.

This is the first version of Chrome that supports CSS Grid without
a flag.

Fixes #2739

Following PR feedback split a parametrized test into three separate
tests for ease of understanding. Also rename `previousSiblingTextLength`
to `previousSiblingsTextLength` to clarify that it includes _all_ of the
previous siblings of the node, not just one.

After resolving an annotation's selectors to a `Range`, convert the
range to a `TextRange` object which is stored with the anchor and later resolved
to a concrete `Range` just before inserting highlights. Representing the
anchored region using a `TextRange` allows for the correct text to be
highlighted later if the region has been modified as a result of
inserting other highlights, providing that the text content of the
region has not been changed.

Fixes https://github.com/hypothesis/h/issues/3278
Fixes https://github.com/hypothesis/client/issues/2326
Fixes https://github.com/hypothesis/h/issues/5997

Add a `TextRange` class which represents a region of the document as an
immutable start/end pair of (element, text position) points. Compared to
a DOM `Range`, this representation of a document region is more robust to
changes in the DOM structure of the region that don't affect the text
content, such as wrapping parts of the range in elements to insert
accessible highlights.

Without the explicit annotation the type of `position` is inferred as
`string` rather than `"relative"|"static"`.

Bumps [preact](https://github.com/preactjs/preact) from 10.5.5 to 10.5.7.
- [Release notes](https://github.com/preactjs/preact/releases)
- [Commits](https://github.com/preactjs/preact/compare/10.5.5...10.5.7)
Signed-off-by: dependabot-preview[bot] <support@dependabot.com>

Bumps [@sentry/browser](https://github.com/getsentry/sentry-javascript) from 5.27.3 to 5.27.4.
- [Release notes](https://github.com/getsentry/sentry-javascript/releases)
- [Changelog](https://github.com/getsentry/sentry-javascript/blob/master/CHANGELOG.md)
- [Commits](https://github.com/getsentry/sentry-javascript/compare/5.27.3...5.27.4)
Signed-off-by: dependabot-preview[bot] <support@dependabot.com>

Bumps [codecov](https://github.com/codecov/codecov-node) from 3.8.0 to 3.8.1.
- [Release notes](https://github.com/codecov/codecov-node/releases)
- [Changelog](https://github.com/codecov/codecov-node/blob/master/CHANGELOG.md)
- [Commits](https://github.com/codecov/codecov-node/compare/v3.8.0...v3.8.1)
Signed-off-by: dependabot-preview[bot] <support@dependabot.com>

Bumps [autoprefixer](https://github.com/postcss/autoprefixer) from 10.0.1 to 10.0.2.
- [Release notes](https://github.com/postcss/autoprefixer/releases)
- [Changelog](https://github.com/postcss/autoprefixer/blob/master/CHANGELOG.md)
- [Commits](https://github.com/postcss/autoprefixer/compare/10.0.1...10.0.2)
Signed-off-by: dependabot-preview[bot] <support@dependabot.com>

Bumps [postcss](https://github.com/postcss/postcss) from 8.1.6 to 8.1.7.
- [Release notes](https://github.com/postcss/postcss/releases)
- [Changelog](https://github.com/postcss/postcss/blob/master/CHANGELOG.md)
- [Commits](https://github.com/postcss/postcss/compare/8.1.6...8.1.7)
Signed-off-by: dependabot-preview[bot] <support@dependabot.com>

Bumps [aws-sdk](https://github.com/aws/aws-sdk-js) from 2.787.0 to 2.792.0.
- [Release notes](https://github.com/aws/aws-sdk-js/releases)
- [Changelog](https://github.com/aws/aws-sdk-js/blob/master/CHANGELOG.md)
- [Commits](https://github.com/aws/aws-sdk-js/compare/v2.787.0...v2.792.0)
Signed-off-by: dependabot-preview[bot] <support@dependabot.com>

Bumps [eslint](https://github.com/eslint/eslint) from 7.12.1 to 7.13.0.
- [Release notes](https://github.com/eslint/eslint/releases)
- [Changelog](https://github.com/eslint/eslint/blob/master/CHANGELOG.md)
- [Commits](https://github.com/eslint/eslint/compare/v7.12.1...v7.13.0)
Signed-off-by: dependabot-preview[bot] <support@dependabot.com>

Fix a bug that could cause pieces of text to be highlighted twice rather
than once. For this bug to occur:

 1. The anchored range had to start in the middle of a text node
 2. The anchored range had to end _after_ the text node, so that the
    range's `commonAncestorContainer` property pointed to an ancestor of
    the text node

In this scenario, the `wholeTextNodesInRange` function would split the
text node where the range started and add the piece on the right-hand
side of the split to the list of text nodes to wrap in highlight
elements. The `forEachNodeInRange` iterator would then visit the
right-hand text node a _second_ time because it is the _next node_ it
its iteration order. This resulted in the text node being added twice to
the `textNodes` list returned by `wholeTextNodesInRange`.

To resolve this, rewrite the loop in `wholeTextNodesInRange` so that it
visits each text node in the range only once, even if the node is split
mid-iteration. In doing so a small optimization was made by setting the
`NodeIterator`'s filter to `NodeFilter.SHOW_TEXT` in order to skip any
non text nodes.

Handle and rebroadcast events in Sidebar to support
the opening and closing of the Notebook. Show
option in user menu if feature flag is enabled.

Add env NOTEBOOK_APP_URL and support for it in
build and config for annotator.

Make some small improvements in response to PR feedback.

Add additional tests for `highlightRange` calls that requires splitting
text nodes in order to wrap only the part inside the range. This turned
up an edge case with handling collapsed ranges that start (and end) in
the middle of a text node.

Decouple `highlightRange` from the `NormalizedRange` class and make it
work with ordinary DOM ranges. This decouples the highlighter
implementation from the range wrapper types, which will make some future changes
to highlighting easier.

`isNodeInRange` in range would throw an exception if the passed node had
no parent because `range.selectNode(...)` requires its argument to have
a parent.

This commit rewrites `isNodeInRange` to use `range.comparePoint` which
avoids creating a temporary live range and handle exceptions that
`comparePoint` may throw. It also updates the JSDoc to more accurately
describe what the function does.

Only show the relative time-since string for the annotation edited date
if it is different from the relative time-since string for the created
date. If they are the same, just show "edited" without further detail.

Fixes https://github.com/hypothesis/product-backlog/issues/1156

When attempting to create a build of the browser extension using the
local dev client and the production h service I encountered an issue where
resolving the `sidebarAppUrl` setting in `src/boot/index.js` failed.

This happens because the extension does not set this setting in the
sidebar app so the boot script fell back to trying to resolve the URL
template baked into the development client. This in turn failed
because `document.currentScript` is not set in this context.

The `sidebarAppUrl` setting is not actually needed when booting the sidebar app, so
this commit refactors the boot code to avoid generating it in this
context. This is done by moving the responsibility for determining which
part of the client to load from `boot/boot.js` into `boot/index.js` and
only resolving the necessary settings.

Rely exclusively on `yarn`.