Resolve derived titles to plain text

[Renakoni]

Summary

Fixes the clunky titles reported for styled first lines: a draft starting with ***<u>alpha</u>*** bravo charlie delta used to lose its emphasis markers but keep the raw <u> tags, so the home list showed <u>alpha</u> bravo charlie delta. Derived titles now always resolve to plain text — <u>alpha</u> … and ***<u>alpha</u>*** … both title as alpha bravo charlie delta.

Kept deliberately separate from #93: this is a user-visible behavior fix on the web side, while the decomposition series PRs are behavior-preserving by contract.

Change

• New shared stripInlineMarkdown reduction in documentState.ts: keeps link/image text, drops inline HTML tags and emphasis markers, collapses whitespace.
• Applied on both title paths, since the same leak existed one # away: the leading-line heuristic from Name untitled drafts from their first meaningful line #83 and the heading branch (# **Trip** to <u>Paris</u> now titles as Trip to Paris).
• The tag pattern matches raw HTML the way CommonMark does (letter-led tag name, optional attributes, optional self-close), so autolinks like <https://example.com> are treated as content and preserved.
• A heading that reduces to pure markup (# ***) now names nothing and falls through to the display-name/leading-text fallbacks instead of titling the document as a row of asterisks.

Titles are derived, never stored as user input, so existing documents pick up clean titles on their next open/save with no migration.

Testing

• Three new/extended unit cases lock the reported case verbatim (with and without emphasis), tag variants (<mark>, <br> word spacing), autolink preservation, styled-heading reduction, and the pure-markup-heading fall-through. documentState suite 23/23.
• Full unit suite 310/310, ESLint, vue-tsc, production build green.

[Renakoni]

Both review points addressed in the updated commit (also rebased onto current main, so the merge commit is gone and the branch is a single commit again):

1. Literal punctuation preserved. The global [*_~]deletion is gone.stripInlineMarkdown` now only unwraps actual paired delimiters:

• code spans `...`
• asterisk emphasis *{1,3}...*{1,3}
• underscore emphasis with CommonMark's intraword rule — _ flanked by letters/digits ((?<![\p{L}\p{N}])_{1,3}([^_]+)_{1,3}(?![\p{L}\p{N}])) never counts as markup
• GFM strikethrough ~~...~~ only (single ~ is content)

So # file_name → file_name, # 2*3 benchmark → 2*3 benchmark, # foo~bar → foo~bar, and suggested filenames keep their underscores. One behavior note: with paired unwrapping, a marker-only heading like # *** no longer collapses to empty by accident, so the helper now explicitly treats a reduction consisting solely of marker characters as "no title" — the existing # *** → fall-through-to-content test kept passing because of this rule.

2. Inline HTML tag pattern fixed. Now /<\/?[a-z][a-z0-9-]*(?:[\s\/](?:[^<>"']|"[^"]*"|'[^']*')*)?>/gi:

• hyphenated custom elements match (<my-widget>alpha</my-widget> beta → alpha beta)
• quoted attribute values may contain > (<span title="a>b">alpha</span> beta → alpha beta, no leaked b">)
• autolinks still don't match because the character after the tag name must be whitespace, /, or > — <https://…> and <user@…> remain content (covered by tests)

Added the three suggested heading regressions plus snake_case_name, ~~done~~, _emphasized_, the custom-element case, the quoted-attribute case, and an email-autolink case. Full suite 312/312, lint and build clean.