Keep rich core Djot on the fast HTML path

[dereuromark]

Outcome

Complete the allocation-light conversion work from #276 for the representative rich core workload, and reduce duplicate document scans on authoritative fallback.

The 49,540-byte carve-bench Djot document now stays on the borrowed path with byte-identical 98,922-byte HTML. It covers headings and sections, title-shaped definition prose, paragraphs, nested strong syntax, unresolved explicit references, tight lists with continuation-shaped nested markers, block quotes, fenced code, and plain-cell tables.

Performance

PHP 8.5.9 NTS, tracing JIT, 20 warmups, fastest of five trials with 50 iterations each:

route	before	this PR	change
rich default facade, 49,540 bytes	13.63 ms / 3.47 MB/s	3.25 ms / 14.52 MB/s	4.19x faster
authoritative owned AST	~13.0 ms / ~3.6 MB/s	10.27 ms / 4.60 MB/s	~21% lower

The facade result is now in the same performance band as carve-php on the publication workload (currently documented at 18.41 MB/s), while emitting almost twice the source size as HTML.

How

• Add conservative borrowed renderers for the exact Djot list, quote, and table shapes whose output can be proven locally.
• Support Djot double-strong nesting, explicit unresolved references, smart quote/dash output, and title-shaped definition prose without AST allocation.
• Retain whole-document fail-closed behavior for real table separators, nested block structure, implicit heading references, ambiguous definitions, and unsupported syntax.
• Gate reference, footnote, abbreviation, and implicit-heading collectors by source-family predicates.
• Skip heading reference extraction and the post-parse heading rewrite when neither implicit heading references nor warning validation can consume them.
• Avoid statistics bookkeeping unless observation is requested.

Correctness

• The complete rich benchmark source is byte-identical to the owned AST oracle.
• Every newly admitted official corpus document is byte-identical; 19 admitted, zero mismatches.
• Focused coverage includes rich list/quote/table shapes, nested strong, unresolved references, title-shaped prose, conventional-table fallback, lazy headings, Unicode, unsafe URLs, and custom renderers.
• Full suite: 2,818 tests, 18,469 assertions.
• PHPStan and PHPCS pass for all changed files.

Validation

vendor/bin/phpunit
vendor/bin/phpstan analyse src/Performance/BorrowedHtmlLayout.php src/Parser/BlockParser.php tests/BorrowedHtmlLayoutTest.php --no-progress
vendor/bin/phpcs src/Performance/BorrowedHtmlLayout.php src/Parser/BlockParser.php tests/BorrowedHtmlLayoutTest.php

After this merges, carve-bench will be refreshed against the new dev-main head as a separate final step.

[codecov]

Codecov Report

❌ Patch coverage is 93.03797% with 11 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.18%. Comparing base (bb14c2a) to head (abed920).

Files with missing lines	Patch %	Lines
src/Performance/BorrowedHtmlLayout.php	92.19%	11 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #277      +/-   ##
============================================
- Coverage     92.26%   92.18%   -0.09%     
- Complexity     3896     3955      +59     
============================================
  Files           111      111              
  Lines         10923    11052     +129     
============================================
+ Hits          10078    10188     +110     
- Misses          845      864      +19     

☔ View full report in Codecov by Harness.
📢 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.