Match djot.js on five parser edge cases (djot.js PR 137 test corpus)

[dereuromark]

Ports the test cases from the pending djot.js PR More test case suggestions from cdjot regressions into the official test suite and fixes the five cases where djot-php diverged from djot.js. All expected outputs were verified against djot.js main (commit 84a4b02), not just taken from the PR.

Fixes

• Invalid attribute specs render literally. Identifiers, class names, keys, and unquoted values only allow ASCII alphanumerics plus _, :, -. Any other character (e.g. {#a<b}) invalidates the entire spec, which stays literal text instead of leaking into attribute context. Applied on the block path, the inline word-attachment path, and trailing attribute blocks after spans/links/emphasis/code. New AttributeParser::isValid() implements the whole-spec grammar check.
• Bare : starts a definition list. A : followed by end of input produces an empty <dl><dt></dt><dd></dd></dl> instead of a paragraph (fuzz-crash class found in cdjot).
• Trailing backslash at end of content is a hard break. para\ at end of input now emits <br> instead of a literal backslash. Also holds at end of a paragraph mid-document and in headings, matching djot.js.
• Bare # marker line continues a heading. # h / # / # x is one heading with content h/x (id h-x), not two sections.
• Table separator cells must start immediately after |. | --- | --- | (leading space) is an ordinary data row; only |---|---| promotes the previous row to a header. Trailing space before the next | remains allowed, matching djot.js pattRowSep.

One existing expectation was corrected: AttributeParserTest previously asserted that an invalid unquoted value (invalid=foo.bar) is skipped while keeping the remaining attributes; djot.js invalidates the whole spec, so the test now expects literal output.

Notes

• Where djot-php deliberately keeps a bracketed run fully literal next to an invalid attribute block (the documented [x]{???} convention), that convention is preserved: [x]{#a<b} renders fully literal, while djot.js keeps the span and renders only the braces literally.
• Two adjacent divergences surfaced while testing but are pre-existing and out of scope here: table cells trim trailing backslash-space differently than djot.js (| a\ |), and an empty higher-level heading between two headings sections differently (# h / ## / # x).

[codecov]

Codecov Report

❌ Patch coverage is 97.50000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 92.39%. Comparing base (2cb432a) to head (c366e36).

Files with missing lines	Patch %	Lines
src/Parser/BlockParser.php	93.33%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #264      +/-   ##
============================================
+ Coverage     92.37%   92.39%   +0.02%     
- Complexity     3660     3668       +8     
============================================
  Files           109      109              
  Lines         10379    10407      +28     
============================================
+ Hits           9588     9616      +28     
  Misses          791      791              

☔ 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.