fix(remend): guard incomplete HTML tag handler against math blocks

[torsello]

Description

handleIncompleteHtmlTag guarded against code blocks but never against math, so an ordinary
comparison inside a math expression matched the incomplete-tag pattern and deleted everything
from the < to the end of the string.

remend('Intro\n\n$$\nI = \\sum_{j<k} p_j\n$$\n\nTAIL')
// before → 'Intro\n\n$$\nI = \\sum_{j\n$$'   ← TAIL gone, math truncated
// after  → unchanged

Because the katex handler (priority 70) runs after htmlTags (priority 10), the now-orphaned
$$ was auto-closed, so KaTeX received unterminated input and rendered a ParseError in
--color-muted-foreground. The net effect for the reader is a message that silently ends
mid-formula in grey text, with no error surfaced anywhere.

remend already implements and exports isWithinMathBlock, and the emphasis handlers already
use it — htmlTags was simply never covered.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Performance improvement
• Refactoring (no functional changes)

Related Issues

Related to #616

_{Deliberately not Fixes — this covers the math half of that issue. See Additional Notes.}

Changes Made

• Added an isWithinMathBlock guard to handleIncompleteHtmlTag, gated behind the same
  hasMathDelimiters fast path the emphasis handlers use, so the O(n) scan is skipped
  entirely for text with no math delimiters.

• Changed the handler to walk forward through candidates instead of bailing out on the
  first guarded one. This is the non-obvious half. incompleteHtmlTagPattern is
  leftmost-matching and its [^>]*$ tail means the match always runs to the end of the
  string — so every later < that starts a tag name is an equally valid candidate. Simply
  returning text when the first match falls inside math would trade one bug for another:

  remend('$x<y$ then <div')
  // guard-only → '$x<y$ then <div'   ← the genuine incomplete tag now survives. wrong.
  // this PR    → '$x<y$ then'        ← correct

  The same walk-forward now applies to code blocks, which fixes a smaller unreported case
  on its own: a < inside a fenced block no longer masks a real incomplete tag after it.

• Added 4 test cases to packages/remend/__tests__/html-tags.test.ts.

• Added a patch changeset for remend.

Testing

• All existing tests pass
• Added new tests for the changes
• Manually tested the changes

New cases cover block $$, inline $, both LaTeX delimiter forms (\(…\) and \[…\]), and
the walk-forward behaviour. I confirmed all four fail on main before the fix and pass after —
the walk-forward case in particular fails against a guard-only implementation too.

Manual verification against the exact reproductions from the issue:

Input	Before	After
Intro\n\n$$\nI = \sum_{j<k} p_j\n$$\n\nTAIL	truncated at \sum_{j	unchanged
inline $$A_{j<k}$$ more\n\nTAIL	inline $$A_{j$$	unchanged
Hello <div	Hello	Hello (unchanged behaviour)

Test Coverage

remend: 493 tests across 26 files, all passing.
Full monorepo suite: 1172 tests across 85 files, 6/6 turbo tasks green.
biome check reports no findings on both touched files.

_{Note: tsc --noEmit reports two TS18048 errors in packages/remend/src/strikethrough-handler.ts.
I verified these are present on main and unrelated to this change.}

Screenshots/Demos

Not applicable — no visual surface changes. The before/after is shown as code above.

Checklist

• My code follows the project's code style
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings or errors
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have created a changeset (pnpm changeset)

_{Documentation left unchecked: this is a behavioural fix with no public API change, so there is
nothing in the docs to update. Happy to add a note to the termination docs if you'd like one.}

Changeset

• I have created a changeset for these changes

.changeset/html-tag-math-guard.md — patch bump for remend.

Additional Notes

What this PR intentionally leaves out. #616 raises a second, broader problem: the handler
also fires on text that is already complete, so ordinary prose like the loop runs while i<n
still loses its remainder. That one is not a missing guard — it needs an explicit signal,
because remend(text) currently cannot distinguish a streaming tail from settled text, and
every tail-oriented heuristic keeps firing after the stream ends.

The issue suggests an option or a mode argument mirroring Streamdown's mode="static".
That is a public API decision affecting more than this handler, so I did not want to make it
unilaterally in a first contribution. Happy to open a follow-up if you want to shape the API —
or to fold it into this PR if you already know which direction you'd prefer.

[vercel]

@torsello is attempting to deploy a commit to the Vercel Team on Vercel.

A member of the Team first needs to authorize it.