fix: scope title:/type: extraction to the frontmatter block, not the whole file

[chuenchen309]

Problem

extractTitle()/extractType() in src/lib/wiki-graph.ts match title:/type: with:

content.match(/^---\n[\s\S]*?^title:\s*["']?(.+?)["']?\s*$/m)

This is anchored at the opening ---, but the lazy [\s\S]*? is never required to stop at the closing ---. A page whose frontmatter has no title/type — but whose body happens to contain a later line starting with title: or type: (plain prose, not YAML) — gets that body line misread as the frontmatter value.

The more severe case: a misread type of "query" matches HIDDEN_TYPES, silently dropping the entire page from the wiki graph.

graph-relevance.ts's extractFrontmatter() already handles this correctly — it isolates the ---...--- block first (/^---\n([\s\S]*?)\n---/) before searching it for title:/type:.

Fix

Extract the same isolated frontmatter block via a small shared helper, and search only within it for both functions.

Testing

Added src/lib/__tests__/wiki-graph.test.ts with two regression tests (mocking @/commands/fs):

• A page with type: entity frontmatter (no title) and a body line title: not-frontmatter-at-all — confirms the label still falls back to the # Real Heading, not the body line.
• A page with title: Real Page frontmatter (no type) and a body line type: query — confirms the page isn't misclassified and dropped from the graph.

Both fail against the pre-fix code and pass after. Ran the full npm run test:mocks suite (1640 passed across 114 files — no regressions) and npm run typecheck (clean).

Disclosure: I used an AI coding assistant (Claude) to help identify this bug and draft the fix. I independently traced the misread-and-drop mechanism through HIDDEN_TYPES, wrote/verified the regression tests in both red and green states, and ran the full local test suite before opening this PR.

[chuenchen309]

Thanks for merging this. I've really enjoyed digging into the frontmatter / wiki-cleanup part of the codebase — I also left reviews on a couple of the open frontmatter PRs (#517 and #561) in case they're useful. You've got a big incoming-PR and issue queue at the moment; if an extra set of hands on reviewing PRs or triaging issues in those areas would help, I'd be glad to pitch in. Entirely your call.