Skip HTML comments in ISO clause structure linter

[OleksiyPuzikov]

The clause linter treated Pandoc RawBlock(html) nodes — which represent HTML comments — as body content between headings, causing false positive violations. Now _is_html_comment_block() detects these blocks (including sourcepos Div wrappers) and excludes them from both the body_seen flag and the body_lines reported in violations.

Attn @asluk - this covers my comment in https://github.com/aousd/core-spec-wg/pull/517

[asluk]

Logic looks correct and conservatively safe (the all() over the unwrapped Div means a comment sharing a block with real prose still flags, so this can't mask a genuine hanging paragraph). One nit:

_is_html_comment_line() is defined and unit-tested but never called in check_file — the body_lines comprehension uses _HTML_COMMENT_RE.match instead. Since that regex only matches a complete single-line comment, a block containing a multi-line comment plus real body text would still leak the comment's <!-- / --> fragments into the violation's displayed context. Cosmetic only (never a false positive — body_seen is governed by the block-level check), but it leaves _is_html_comment_line as dead code. Either wire it into that comprehension or drop the helper.

[OleksiyPuzikov]

@asluk updated.

[asluk]

tl;dr the rewrite addresses my earlier nit and it looks good to merge. Pulled the head (b09cab7) and verified locally:

• All 24 tests pass (python -m unittest tests.test_iso_clause_lint).
• Run against specification/ drops the count from 114 to 111: the three removed are top-level headings whose only body was a comment (Compliance.md:1, Scope.md:1 (the D7 <!-- ISO TODO --> placeholder), Versioning.md:1). No genuine hanging paragraph is masked, and no surviving violation has a comment line left in its reported body.
• Clean against current main: ahead 3, behind 0, touching only iso_clause_lint.py and the new test file.

The two-helper split reads well: _is_html_comment_block gating body_seen for the count, _strip_html_comment_lines cleaning the displayed body. Thanks for the thorough test coverage.

This isn't gating #517 (that pin stays at c918081), so it can ride a later routine pin bump. If you consider it done, mind taking it out of draft so we can merge?