This commit has been cherry picked from its original appearance in enketo#164, with the following changes:

- Include all enketo-core fixtures in preparation for adding benchmarks to establish a performance baseline.
- Temporarily increase the timeout for snapshot tests so that the longer running enketo-core fixtures can generate initial snapshots.
- Restore test shuffling (turning that off was a mistake in the original TypeScript PR).
- Switch the default test reporter to `dot` to improve visibility of progress during test runs (it's still not great, mainly due to performance).
- Move `linkedom` stuff to `/test/shared.ts` as it will only be used in tests.
- Accept updates to `package-lock.json` with SHA-512 integrity hashes.
- Excludes formatting of XML/XSL.

About snapshot serialization:

Snapshots use a custom serializer designed to identify only meaningful differences. This means that a snapshot matches if:

- the value is exactly equal
- the value is equal, apart from insignificant changes to whitespace and/or attribute order

The serialization logic:

1. Manually ensures consistent attribute order (using `linkedom` as a DOM compatibility library, primarily because its types are much better than those for `@xmldom/xmldom`).
2. Uses `prettier` to normalize whitespace.
3. Uses `pretty-format` (`vitest`'s default) for final serialization.

Apart from adding snapshots, tests have been updated to reference `linkedom`'s types (corrected where necessary) and to its behavior (which is semantically equivalent to the previous `@xmldom/xmldom` usage, but more consistent with a real DOM environment).

This is almost entirely different from the original in enketo#164. Instead it:

- uses benchmark.js rather than implementing our own
- mostly defers to benchmark.js stats logic for reporting
- skips `@actions/core` for generating the GitHub Actions summary, it wasn't really doing a lot for us

This should have little impact on forms which already transform quickly, but it dramatically improves the time to transform forms which are particularly slow. In my testing, this is especially the case for forms with a large number of translations/itext values.

The improvement comes from using `xsl:key` elements to reference the expressions identified as expensive, and to look them up with the corresponding `key` XPath function. My strong assumption is that the reason this has such a big impact is that the non-key references are rerun on each visit to their respective `xsl:template` or upon each reference to a top-level `xsl:variable`, whereas `key` presumably pre-caches the expression's nodeset and only filters based on the `use` argument.

…json version

Otherwise snapshots will not match as soon as we prepare release

This has been validated locally in enketo-core and enketo-transformer using `npm link`.