…est and CI job

Combine test/html5lib-webpack.spectest.js and
test/html5lib-tree-construction.spectest.js into a single
test/html5lib.spectest.js with two describe blocks, and drop the separate
html5lib-tree CI job and test:/cover: scripts. The single html5lib CI job
now runs both the tokenizer no-crash builds and the tree-construction
conformance checks.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

The change ships no user-facing release note, so use an empty changeset
(no package bump) instead of a patch entry.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

Resolve 16 of the documented conformance divergences:

- EOF-truncated tags: in rawtext "text" mode, an end tag whose name ran to
  EOF with no delimiter is rawtext, not a close; drop start tags the tokenizer
  only emitted because it hit EOF mid-tag (no closing `>`).
- `select`: remove the now-dead "in select"/"in select in table" insertion
  modes and route `select` content (including fragment contexts) through
  "in body", matching the spec's removal of the select insertion modes.
- `colgroup`: split a character token's leading whitespace in "in column
  group" so non-whitespace fosters out while the whitespace stays.
- `noscript`: ignore a stray end tag in "in head noscript" instead of
  popping <noscript>.

The remaining 15 are documented in the spec test: the foreign-content
breakout for stray `</p>`/`</br>`, `<selectedcontent>` mirroring, EOF inside
a quoted attribute value, and `<input>` in a select fragment context.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

Drop the changeset entirely so the change ships with no release note.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

- Foreign-content breakout: special-case stray `</p>`/`</br>` end tags to pop
  foreign elements up to the nearest HTML element or integration point, then
  process in "in body" (matching the reference algorithm). Fixes the 8
  foreign-breakout cases without regressing the scope-based "ignore stray end
  tag" cases.
- EOF mid-tag: drop a start tag the tokenizer only emitted on an eof-in-tag
  parse error (covers EOF inside a quoted attribute value), via a parseError
  hook instead of a `>` heuristic.
- `<input>` in a select context is dropped (closing an open select first);
  keygen/textarea keep their normal behavior.

Only the brand-new `<selectedcontent>` cases remain divergent (not modelled
by the reference parser either); they stay documented in the spec test.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

Fix the last divergences so buildHtmlAst matches every tree-construction
case (scripting-enabled cases excepted):

- `<select>` inserts an active-formatting marker so a stray formatting end
  tag (e.g. `</font>`) cannot adopt across the select boundary now that
  select has no dedicated insertion mode.
- `<selectedcontent>` mirrors its `<select>`'s selected `<option>` subtree: a
  gated post-parse pass deep-clones the selected option (last `selected`,
  else first) into each `<selectedcontent>`, with source offsets stripped so
  no duplicate dependencies are emitted.

KNOWN_DIVERGENCES is now empty; the spec test asserts an exact match for all
1783 cases.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

The tree builder only ever ran on walkHtmlTokens' token stream (and drove it
through the isForeign/fragmentContext hooks), so fold it into that module:

- Move the AST builder, its typedefs and helpers into walkHtmlTokens.js and
  expose `walkHtmlTokens.buildAst(source, fragmentContext)` plus the
  `NS_HTML`/`NS_MATHML`/`NS_SVG` constants; delete lib/html/buildHtmlAst.js.
- HtmlParser now calls `walkHtmlTokens.buildAst` and imports the AST typedefs
  from `./walkHtmlTokens`.
- Update the unit/spec tests to the new entry point (rename the AST unit test
  to walkHtmlTokens.buildAst.unittest.js; drop the now-obsolete tokenizer
  mock in favor of a real foster-parenting text-merge case).

No behavior change; html5lib tree-construction stays at full conformance.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

Make ES-module handling of `<script>` configurable instead of being derived
solely from `output.module`:

- New `module.parser.html.scriptModule` boolean (default `false`). When on, a
  `<script src>` / inline `<script>` without an explicit `type` is treated as
  an ES module script (tag emitted as `type="module"`, entry bundled as ESM).
  `<script type="module">` is always a module; classic by default matches the
  browser/other-bundler convention.
- Effective module-ness is `scriptModule || output.module`, so ES-module
  output still emits module scripts (the script entry's chunk format follows
  `output.module`, so the tag must match) and the option only adds opt-in.
- Schema + defaults wired; HtmlParser reads it and replaces the hardcoded
  `output.module` checks for script-type decisions.

Adds a configCases/html/script-module integration test (classic output +
scriptModule:true upgrades typeless scripts) and refreshes the Defaults /
CLI-args snapshots.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

- Add walkHtmlTokens.buildAst unit tests for paths the html5lib corpus
  doesn't hit: `<dd>`/`<dt>` auto-close + end tags, quirks mode from a
  4.01-Transitional doctype (`<table>` stays inside `<p>`), `<selectedcontent>`
  mirroring (incl. cloning an attributed child with stripped offsets and the
  last `selected` option), and fostering stray text in a `table` fragment
  context.
- Remove the dead `scriptingFlag` branches: scripting is always disabled, so
  the noscript-when-scripting paths and the fragment ternary were unreachable.

walkHtmlTokens.js reaches 100% functions / ~99.7% lines, ~97.9% branches.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

Remove the `module.parser.html.scriptModule` parser option for now; it will be
implemented later. Script module-ness is again derived solely from
`output.module`, as before.

This reverts commit c8f0095, restoring the schema, defaults, HtmlParser logic,
generated declarations/types, the Defaults/CLI snapshots, and dropping the
script-module integration test.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5

Apply the spec-conformant tree-builder fixes directly in lib/html/buildHtmlAst.js
instead of folding it into walkHtmlTokens, so walkHtmlTokens stays tokenizer-only
and the diff stays focused.

https://claude.ai/code/session_0156ZAChgvF8kUGMh6Qihsy5