test(e2e): cover useFocusTrap's previousFocus cleanup fallback when the trigger unmounts

[hivecommons-hive]

Test Improvement

Adds one new end-to-end spec, tests/e2e/focus-trap-trigger-unmounted.spec.js,
for the only close path that nulls a lightbox's trigger ref.

src/components/hooks/useFocusTrap.js:51 restores focus with
(triggerRef.current || previousFocus)?.focus?.(). Every existing lightbox spec
closes the dialog with Escape, the close button or a backdrop click, and in all
three the trigger is still mounted, so the previousFocus operand is never
evaluated in a browser.

Both lightboxes are rendered by the card that triggers them
(src/components/MemberDirectory/MemberCard.js:72-78), and the directory's
search field stays live behind the backdrop. A visitor who opens a profile and
keeps typing filters that card out: trigger and dialog unmount in the same
commit, React detaches refs before running passive-effect cleanups, and the
cleanup falls through to previousFocus.

The spec drives exactly that: open a profile, focus #member-search through the
DOM (a click would dismiss via the backdrop while the trigger is still mounted),
type a query no organization matches, then assert

• the dialog unmounts with its card,
• no pageerror fires — written as triggerRef.current.focus() the cleanup
  throws a TypeError into the commit while the visitor is typing,
• document.body.style.overflow returns to the value captured before the
  dialog opened — a cleanup that throws early leaves the page unscrollable with
  no dialog on screen,
• focus stays in the search field with its typed value intact,
• and the directory is still live after the query is cleared, so the assertions
  describe a working page rather than one that stopped responding.

Verification at 03cfcfe

• Full two-build coverage run (npm run build:e2e:coverage, then
  test:e2e:coverage): 341 passed, and the gated report
  (--check-source 100 --check-source-regions 91 --require-source-files) exits
  0 at src files | 100.00 | 91.72 | 2099/2099 lines | 443/483 regions, up
  from 91.63% / 438 of 478 on main.
• npm run test:unit: 2025 passed, 0 failed. npm run check:format: clean.

This does not retire the region, and that is #1066 / #1079

Stated up front rather than claimed otherwise. A control run of this spec alone
against the real build reports useFocusTrap's uncovered regions as 29 34 35
— line 51 absent, i.e. demonstrably executed. The full two-build run puts it
back (35 51), because the variant build emits the cleanup under coordinates
the real build's key does not match. That is the region-union attribution
behaviour tracked by #1066 and #1079, and this spec is written as behaviour
rather than as a coverage claim for the same reason
tests/e2e/dialog-non-dismissing.spec.js already is.

Line 35 (if (!focusable?.length) return;) is deliberately out of scope: both
lightboxes always render a close button, so the taken arm has no browser path
at all.

Files

One new file, nothing edited:

• tests/e2e/focus-trap-trigger-unmounted.spec.js

Disjoint from every open hold-gated PR — not scripts/audit-gate.mjs (#1161),
tests/tools/e2e-coverage-report.mjs (#1163), scripts/lib/svg-active-content.mjs
(#1168, #1176), CONTRIBUTING.md (#1166), tests/architecture-content-mirror.test.mjs
(#1165), tests/tools/e2e-data-fixtures.cjs or the metrics.json overlay
(#1171), or tests/e2e-coverage-run.test.mjs (#1174). No fixture overlay is
added, so the two-build ceiling in #1172 is untouched, and no workflow file is
involved.

Related Issue

Closes #1177

---

Human review required; this PR carries hold and must not be merged by automation.

— hive: agent=quality backend=copilot model=claude-opus-5 copilot=1.0.88

[hivecommons-hive]

Important

Held for human review by the hive's ACMM level gate.

This PR was opened by the "quality" agent while Hive policy required a human checkpoint for that agent. Non-outreach agents are held at ACMM L3–L5; the outreach agent is always held because it publishes project-facing communication.

Hive will keep the hold label until a human removes it. Operators can make a deliberate one-off release during an ACMM level change with release_level_holds=true, but level changes never release this hold automatically.