Title every appendix section, localized, and drop the footnotes rule (bd-v9zs83zj)

[cscheid]

Closes bd-v9zs83zj.

q2's footnotes appendix had no visible heading. appendix.rs pushed the extracted
footnotes Div verbatim, where the branch directly above it calls
wrap_bibliography. That asymmetry is the bug in one sentence; this adds the missing
wrap_footnotes.

Two things widened the fix beyond the strand's title, both agreed before implementation.

1. The <hr> goes too — the strand's premise was wrong here

The strand states the <hr /> is correct and only the heading is missing. It isn't.
Quarto 1's prependHeading (format-html-shared.ts:405-409) inserts the heading and
removes the rule — the heading takes over as the section separator. Shipping the
heading alone would leave q2 emitting both, and the Connect-docs diff this strand exists
to close would not have closed.

The strip lives in the appendix transform, not in create_footnotes_section, so the
rule survives when appendix processing is off (appendix-style: none, book: true).
That matches Q1, which only titles footnotes from inside processDocumentAppendix.

2. All five headings, not one

References / Reuse / Copyright / Citation were hardcoded English despite
section-title-* keys existing in _language.yml and being fully translated across
every _language-*.yml. They now route through the same localized lookup, following
toc_generate.rs's precedence (localized term > English literal — the literal covering
stage-less unit tests, where LanguageTerms::from_meta returns None).

Heading shape

class="anchored quarto-appendix-heading", no id — matching Q1 exactly.

• quarto-appendix-heading activates SCSS q2 already shipped but never matched
  (_bootstrap-rules.scss). This is a real visual change.
• anchored is inert in q2: in Q1 it is an AnchorJS selector hook, and q2 ships
  neither the runtime nor any rule for it. Emitted so the appendix is already correct
  when heading anchors land — follow-up bd-5kf2dnw4, which must skip headings that
  already carry the class, since Vec<String> push is not idempotent the way
  classList.add is.

Titles split into canonical Str/Space inlines rather than one Str with embedded
spaces. Never observable in English (every title is one word); required once localized
(Derechos de autor).

Review notes

The two must-not-change cases — appendix-style: none and book: true — passed
before and after the change. They are the guard on the conditional <hr> strip.

appendix_title filters an empty resolved term before falling back: a language term set
to null round-trips as Some(""), which would emit a heading with no text.

Not using pampa's split_string_to_inlines — it is Lua-module-scoped and hardcodes
By::unknown() provenance; the local helper carries By::appendix().

Testing

7 new unit tests in appendix.rs + 3 smoke-all fixtures, all confirmed failing
first (6 unit failures, 7 regex mismatches).

Snapshots: zero changed. Verified this is real coverage rather than a blind spot —
no .snap in the tree contains appendix HTML at all (quarto-appendix /
doc-endnotes / quarto-bibliography all return nothing). The appendix is covered by
smoke-all and unit tests instead.

Full cargo xtask verify (not --skip-hub-build): 14/14 steps, 11801 passed, 197
skipped.

End-to-end, three ways:

1. cargo run --bin q2 -- render → <h2 class="anchored quarto-appendix-heading">Footnotes</h2>,
   no <hr>. Byte-identical to Q1.
2. Computed styles read back from a real browser page: font-weight 600,
   font-size 17px (= 1em), margins 0, border-bottom none, opacity .9 — every
   rule in the previously-dead block now applies. #footnotes hr → 0.
3. The user/manifest Connect-docs page re-rendered and compared against the Q1
   reference site: previously 0 occurrences of quarto-appendix-heading, now matching
   exactly.

Design rationale and the full investigation are in
claude-notes/plans/2026-08-12-footnotes-appendix-heading.md.

Unrelated, filed separately

The first full verify failed at step 4 on a tree-sitter corpus case this change never
touches. Chased down: it reproduces on pristine main with the tree stashed,
tree-sitter generate yields no diff (so parser.c is in sync with grammar.js), and
it passes 3/3 after a forced rebuild. The culprit is the stale gitignored
markdown.dylib build cache. Filed as bd-7ilvb5r2; not a grammar bug and not this
change.

🤖 Generated with Claude Code

[posit-snyk-bot]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.