Use CommonMark parsing for section-header-quality Markdown headings

[dacharyc]

Purpose

Follow-up to merged PR #150, which introduced CommonMark parsing with GFM table support. Replace the Markdown heading regex in section-header-quality with parser-backed heading extraction while preserving existing HTML/MDX and callout behavior.

Recommended sequence: reuse any shared parser helper established by #151. This is a separate bounded migration and does not depend on completing the content-parity migration.

Current Implementation and Risks

src/checks/content-structure/section-header-quality.ts uses extractHeaders() to combine HTML headings from node-html-parser with Markdown matches from MD_HEADING_RE.

The Markdown pass scans raw source for ATX headings. It does not recognize setext headings or distinguish a heading from matching text inside a code example. It also leaves Markdown inline formatting in the extracted heading text. Add reproductions before changing behavior; these risks were identified by inspection, unlike the independently reproduced link-extraction bugs tracked in #151.

The HTML path intentionally excludes headings in callout/admonition containers using ancestor tags, ARIA roles, classes, and data attributes. That behavior fixed #51 and must be preserved.

Implementation Surface

• src/checks/content-structure/section-header-quality.ts: extractHeaders(), MD_HEADING_RE, and isCalloutHeading().
• test/unit/checks/section-header-quality.test.ts: extend existing check-level fixtures.
• Inspect the tabbed-content-serialization panel contract and HTML/MDX examples before deciding how parsed Markdown and HTML headings are combined.

Acceptance Criteria

• Add focused regressions for missed setext headings, ATX-looking code/comment examples, closing ATX hashes, and formatted heading labels.
• Extract heading text from actual Markdown heading nodes. Handle emphasis, links, inline code, escaped punctuation, and entities according to rendered-text semantics.
• Exclude heading-like text in fenced/indented code and HTML comments, including nested containers and indented comments. Verify LF/CRLF behavior.
• Preserve HTML heading extraction and the callout/admonition exclusions from Callout headings trigger "section-header-quality: tab headers don't distinguish between variants" #51. Respect ancestor context in mixed HTML/Markdown or MDX panels; a second extraction path must not reintroduce excluded headings or count a heading twice.
• Test equivalent HTML and Markdown panels and the intended outcomes for contextual versus repeated generic headers, including cross-group comparisons.
• Preserve check dependencies, public details/results, classification thresholds, and request behavior. Keep the work limited to accurately extracting header text and structure.
• Remove the replaced Markdown heading regex, reuse shared parser infrastructure where appropriate, and run affected tests plus repository CI gates.

Boundaries

Do not replace HTML parsing with CommonMark, introduce a wholesale MDX migration, or redefine which supplementary callout headings count as structural headers. Do not migrate content-start detection, soft-404 heuristics, llms.txt structural validation, link extraction, content parity, or fence validity in this issue.

The parser determines Markdown structure; the existing check still decides whether a section label distinguishes tab variants. No implementation is included in this issue.