Test all CommonMark spec examples

[bobzhang]

Split out of #152. Merge last: this requires all the other PRs split out of #152. On main alone, this test fails with 631/652, so CI stays red until they are merged.

Adds src/cmark_html/spec_examples_test.mbt. It renders every example of src/data/test/spec.md (CommonMark 0.31.2) with the XHTML renderer, compares the result to the expected HTML, and fails if any example fails. The examples are extracted by a small state machine over the lines of the spec.

	passing examples
main	631 / 652
with all PRs below	652 / 652

Required PRs (spec examples each one fixes):

• Match link labels with Unicode case folding #153 link label case folding (206, 540)
• Reject invalid raw HTML tags, attributes and email autolinks #154 invalid raw HTML and autolinks (606, 618; 621 with Fix inline raw HTML spanning lines: restore tokens, CommonMark 0.31 comments, linear-time guard #156)
• Fix inline raw HTML spanning lines: restore tokens, CommonMark 0.31 comments, linear-time guard #156 inline raw HTML spanning lines: restore tokens, 0.31 comments (<!--> / <!--->), linear-time guard (626; 621 with Reject invalid raw HTML tags, attributes and email autolinks #154). Accept <!--> and <!---> as HTML comments #155 was folded into it.
• Count Unicode symbols as punctuation for emphasis flanking #161 Unicode symbols are punctuation (354)
• Reject link reference definitions with text after the title #162 link reference definitions with text after the title (209, 210)
• Concatenate the text segments of image alt text #163 image alt text (520, 573–577, 585, 589)
• Process emphasis with the spec's delimiter stack algorithm #167 process emphasis (411, 412, 415, 429), stacked on Count Unicode symbols as punctuation for emphasis flanking #161

The other split PRs (#157, #158, #159, #160, #164, #165, #166) fix bugs that no spec example covers.

Verified on a local branch that merges all of them: 652/652 on native, js, wasm-gc and wasm. All regression tests of #152 also pass there.

[bobzhang]

Automated review: Codex CLI (codex exec, model reasoning effort high, read-only sandbox) on git diff against the base branch. 2 rounds.

• Round 1: P3, the failing example numbers were never shown because the count assertion failed first. Fixed in 85258c0: the failure list is now asserted first.
• Round 2: signed off, subject to the documented merge order.

I sign off on the full diff, subject to the documented prerequisite PRs merging first. No concrete problems found.

• Round-one assertion-order issue is fixed.
• All 652 extracted Markdown/HTML pairs match the upstream 0.31.2 extractor, including tabs and trailing newlines.
• Strict parsing, XHTML rendering, and per-document renderer state are appropriate.

Codex could not run moon test in its read-only sandbox. The tests were run separately on native, js and wasm-gc and pass.