test(fold): assert the duplicate-title fold behaviour instead of its spelling

[PathGao]

What this is

The first slice of converting source-shape assertions into behaviour tests, following #563. Two test files change; no source file does.

• foldKeys.test.ts — rewritten to run processMarkdownHtml instead of matching markdown.ts with regexes.
• findRefocus.test.ts — deleted.

Mechanism

The old file asserted /h\.id = \w+\.id/ and removeAttribute("id") against markdown.ts. That is a claim about how the promotion is spelled, and it fails in both directions:

	promotion deleted	promotion extracted into a helper
old assertions	red ✓	red ✗ — nothing broke
new assertions	red ✓	green ✓

Both rows were measured, not reasoned about. The refactor in the right-hand column is four lines moved into function promoteAnchorId(h) above processMarkdownHtml; the whole suite stays green and the old assertion still goes red, because the text it was looking for moved.

The old file also could not see the part that matters. It checked that an id gets assigned, never that the ids stay unique afterwards — which is the reason the anchor gives its own up rather than sharing it.

None of this needed new infrastructure. processMarkdownHtml is exported from .ts, and renderProtocolDom.ts (the suite's DOM shim, already used by foldStatePerDocument, exportRichContent, diagramCache and three more) supplies the DOM it wants. The test document now contains the collision the feature exists for — two sections both titled Setup — and asserts that collapsing the first leaves the second open. Before the promotion, keyed by textContent, new Set(['setup']) collapsed both.

Scope

• No source changes. The helper extraction above was an experiment to measure the false-red; it was reverted.
• One source-shape assertion is kept on purpose, with the reason written next to it: the preview's foldableHeader.id || foldableHeader.textContent lives in a Svelte component this runner cannot import, and it is a contract between the producer and its consumers rather than an internal call site.
• findRefocus.test.ts came in with fix(find): re-focus the preview find bar on a repeated Cmd/Ctrl+F #560 and was mine. It pinned findBar?.focusInput() — an internal call, free to rename, and green if the focus stops working. By the criterion docs: name the two kinds of test in scripts/, and ask which one a PR added #563 adds to the template it should not have been written; deleting it is cheaper than defending it.
• The other 30 or so files that assert source shape are untouched. Most of them anchor on a genuine contract (a Tauri command name, a Tauri capability, version sync across package.json/Cargo.toml, a duplicated implementation) and should stay as they are.

Tests

The two behaviour tests are the fix check for themselves: delete the promotion block from markdown.ts and both go red; restore it and they pass. Ran both directions.

Verification

npm run check   # 674 files, 0 errors
npm test        # 942 pass, 0 fail  (944 before: 3 deleted with findRefocus, 4 added here, 1 net)

Not verified: nothing here reaches a browser, so the shim's fidelity to real classList/querySelector behaviour is taken on the same trust the five existing shim-based test files already take it on.

🤖 Generated with Claude Code