Fix HtmlToDjot reverse-conversion round-trip fidelity

[dereuromark]

Problem

The default DjotConverter output (without round-trip annotations) did not survive HtmlToDjot back-conversion in several cases, so a Djot -> HTML -> Djot -> HTML cycle changed the rendered HTML. This is exactly the path WYSIWYG and HTML-serializing tools rely on, where HTML stability across the round-trip matters.

Found via a round-trip drift harness over the core element set plus element combinations.

Fixes

Bracket-in-label corruption. processLink() blanket-escaped every ] in the already-built child markup, mangling structural brackets produced by a nested image or link. [![alt](/i.png)](https://x.com) became [![alt\](/i.png)](https://x.com) (invalid). ] is now escaped at the text-node level in escapeDjotText(), where it is known to be literal, so structural ] emitted by child element processors stays intact. A literal ] inside a link/span label is still neutralized. This also fixes literal ] inside span and semantic-span labels, which had the same latent issue.

Footnote reference. The default footnote reference <a href="#fnN" role="doc-noteref"> was unrecognized and produced invalid Djot ([^1^](#fn1){#fnref1}). It now round-trips to a footnote marker, pairing with the definition already collected by processEndnotesSection(). Annotated inline footnotes (which also carry role="doc-noteref") are explicitly excluded and unaffected.

Loose lists. A loose list (each item wrapped in <p>) collapsed into a tight list on the way back. Items are now separated by a blank line when the list is loose, so the next parse stays loose. Tight lists are unchanged.

Explicit heading and section ids. A <section id="..."> whose id is not derivable from the heading text dropped the id, so the next render generated a different one. Explicit ids are now preserved, while auto-derived slugs stay implicit so they regenerate identically. The unsluggable-heading case (where Djot falls back to an s-N id) is handled: an s-N id is treated as auto, any other id as explicit. Auto-slug detection reuses HeadingIdTracker::normalizeId() so it cannot drift from the renderer (transliteration included).

Not addressed (documented as known limitations)

These remain "drift" in the harness but are not reverse-converter bugs:

• Soft break inside block text normalizes a newline to a space - renders identically.
• A raw HTML block (=html) round-trips to a paragraph - the reverse side cannot know the original was a raw block without annotation.
• <b> maps to strong markup - semantically equivalent.

Notes

Unit tests cover each fix with exact reverse output, plus a new HTML-stability data-provider suite for the default (un-annotated) converter path. Full suite green (2485 tests), phpstan and phpcs clean.

[codecov]

Codecov Report

❌ Patch coverage is 93.93939% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.87%. Comparing base (72200ec) to head (2afb8e0).

Files with missing lines	Patch %	Lines
src/Converter/HtmlToDjot.php	93.93%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff            @@
##             master     #208   +/-   ##
=========================================
  Coverage     91.87%   91.87%           
- Complexity     3470     3489   +19     
=========================================
  Files           104      104           
  Lines          9829     9860   +31     
=========================================
+ Hits           9030     9059   +29     
- Misses          799      801    +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.