This index was used to establish declaration order of beforeEach and 
afterEach hooks. Since these hooks no longer support snapshots, 
associatedTaskIndex is unnecessary.

This index was used to establish declaration order of snapshots with 
ids. Since snapshots with ids are no longer supported, this index is 
unnecessary.

This format contains all the necessary information for regenerating
snapshot reports (test titles, snapshot labels, snapshot ordering) but
isn't designed for robustness of any sort, only ease of implementation.
Once a proof of concept is ready, the format should be revisited.

This requires deleting all the .snap files, as well as modifying 
contains-only.js and then putting it back after.

Still prohibit removing snapshots when any block was skipped.

This enables unit tests to continue using Runners without snapshot 
support, while allowing Runner to use the snapshot manager interface as 
though it was pre-loaded.

Including test.only(), --match, and line number selection.

Better conveys the purpose of the test suite.

Clarifies the PR diff, since the old snapshot-updates is being entirely
removed and replaced.

This reverts commit a777048.

This reverts commit 15e226c.

When new t.snapshot.skip() assertions are followed by a t.snapshot()
assertion, the run will fail with a RangeError, as a result of
attempting to save nonconsecutive snapshots. This can only happen when
the t.snapshot.skip()s are wholly new, not when they are the result of
skipping an existing snapshot assertion.

This reverts commit 98e7123.

Simplifies runner's logic, avoids needing to maintain two separate
copies of the snapshot manager interface.

Avoids some redundant path.join()s.

Relies on stable sorting. Stable sorting is present in Node v12+ and 
required as of ECMAScript 2019.

Fixes edge case with invalid .snap and no snapshot assertions, which 
produces an empty newBlocksByTitle map without --update-snapshots.

Allows changing the default reported label in the future if desired.

This occurs when the old snapshots are unreadable or `t.snapshot.skip()`
is added at the end of a test.

Previously, skipping snapshots when the old snapshots are unreadable
would throw either SnapshotError or 'Can't skip snapshot assertions when
updating snapshots'. Adding `t.snapshot.skip()` to the end of a test
would throw RangeError.

Blanks are formatted as `<No Data>` in snapshot reports.

Blanks compare equal to anything and are filled in when a value is
passed; essentially, they behave like they didn't exist.

Internally, blanks are represented with `{buffer: undefined}`.

Ensures that --update-snapshots can fix cases where the snapshot file is
arbitrarily malformed.

These cases previously did, and still do, throw an unhelpful error when
not run with --update-snapshots.

In snapshot format v2, --update-snapshots did not load the old file at
all, avoiding this issue.

We can't ensure deterministic output, but omitting undefined properties,
requiring preferred serialization, and sorting map keys should make
deterministic output more likely, at the cost of somewhat reduced speed.

Canonical CBOR is an outdated specification, replaced in RFC 8949 by
"Deterministically Encoded CBOR". If `node-cbor` comes to support the
later, updating to it should be backwards-compatible; both are valid
CBOR, the major difference being the map key sort order.

This reverts commit f9e92e3.

Keep docs accurate for AVA 3.

Co-authored-by: Mark Wubben <mark@novemberborn.net>

The blocksByTitle format is an implementation detail of Manager, whereas
the CBOR data structure is a characteristic of the snapshot file
protocol.

This change reduces codepath duplication between generateReport() and
encodeSnapshots(), and avoids passing blockIndices around everywhere.

If decodeSnapshots() and encodeSnapshots() are later exposed as a
library, this may be closer to the desired interface.

Hopefully will make the snapshot-worker snapshots easier to read.