[test] built-in /code-review in CI — do not merge

[morisil]

Throwaway PR for testing whether Claude Code's built-in /code-review --comment works from claude-code-action.

• .github/workflows/claude-code-review.yml points at the experiment branch xemantic/.github@claude-review-builtin-test. That branch runs /code-review medium --comment <PR> with show_full_output: true.
• isHtmlBlank has a deliberate bug (all → any) for the review to find.

Close without merging once the run has been inspected.

🤖 Generated with Claude Code

[github-actions]

Confirmed — this function has multiple real call sites (whitespace collapsing, actionable-element labeling, blank-inline-formatting dropping), so the bug is live and impactful, not just cosmetic.

[
  {
    "file": "markanywhere-html-spec/src/commonMain/kotlin/HtmlWhitespace.kt",
    "line": 43,
    "summary": "`isHtmlBlank()` uses `any` instead of `all`, so it now returns true if a string merely *contains* HTML whitespace anywhere, rather than being *entirely* whitespace — inverting the intended semantics and contradicting its own doc comment (\"every character is HTML whitespace\").",
    "failure_scenario": "For non-blank text with an embedded space/tab/newline, e.g. \"Fundstelle openJur\" or \"§ 823\" (with an ordinary space), `isHtmlBlank()` now returns `true` instead of `false`. Downstream this makes `HtmlWhitespaceNormalization.kt:104` treat real content as blank/pending whitespace (merging or dropping visible text), `LabelActionableElements.kt:207` treat a text node with any space in it as not counting as a label (an `<a>` with normal multi-word text could be judged unlabelled and get a placeholder or be misjudged), and `DropBlankInlineFormatting.kt:69` treat non-blank text as blank and strip formatting/mark pairs that actually have real content — e.g. `<em>Fundstelle openJur</em>` would be dropped as if it were empty, since almost every real sentence contains at least one whitespace character."
  }
]

Note: this is a throwaway/test PR (per its description) with the all→any bug deliberately planted to verify the review pipeline works — the finding above is exactly that intentional bug, correctly caught.

_{Posted by the workflow: the review agent finished without commenting.}

[morisil]

Test done: the built-in /code-review --comment found the planted bug but did not post from CI; the workflow's fallback step posted it. Keeping the plugin-based review.