test: cover the keys useFocusTrap deliberately ignores

[hivecommons-hive]

Test Improvement

Adds tests/e2e/focus-trap-indifferent-keys.spec.js (new file, nothing else
touched): two cases, one per lightbox, that assert
src/components/hooks/useFocusTrap.js ignores a key that is neither Tab nor
Escape.

The hook installs one document-level keydown listener for the lifetime of
an open dialog, so it sees every key pressed anywhere on the page. Its Escape
arm (community-people.spec.js:191, member-directory.spec.js) and both Tab
wrap arms (community-people.spec.js:202, interactions.spec.js:217) are
already driven in a browser. The early return at line 33 was not:

33  if (event.key !== 'Tab') return;

Without it, a keystroke made with focus on the dialog's last focusable
element satisfies !event.shiftKey && document.activeElement === last and pulls
focus back to the close button — typing in a dialog, or arrow-scrolling it,
would bounce focus on every keypress. Each case therefore Shift+Tabs to the last
focusable before pressing its indifferent key, which is the only position from
which the guard's removal is observable, reads focus back as an index into the
same button, a[href] list the trap queries (so the assertion is "focus did not
move", not "something is focused"), and presses Escape afterwards to show the
listener was still attached.

Verification

All run locally at 7ab301e against npm run build:e2e:coverage:

• npx playwright test tests/e2e/focus-trap-indifferent-keys.spec.js — 2 passed.
• Full npm run test:e2e:coverage with the spec in place — 324 passed.
• npm run test:unit — 1966 passed, 0 failed.
• npx prettier --check on the new file — clean.
• Mutation check. Replacing line 33 with a comment and rebuilding, then
  running this spec plus community-people.spec.js, member-directory.spec.js
  and interactions.spec.js: 4 failed, 33 passed. Both new cases fail, as
  do the two pre-existing Shift+Tab wrap cases (the bare Shift keydown now
  falls through and fires the backward wrap before Tab is ever pressed) — which
  is the same defect seen from a different angle, not this guard's own coverage.

Region effect. Two full test:e2e:coverage runs against the same build:

without: src/components/hooks/useFocusTrap.js | 100.00 | 85.71 |  | 33 35 51
with:    src/components/hooks/useFocusTrap.js | 100.00 | 90.48 |  | 35 51
without: src files | 100.00 | 91.45 | 2099/2099 lines | 417/456 regions
with:    src files | 100.00 | 91.61 | 2099/2099 lines | 415/453 regions

Unlike #1097 and #1120, region 33 does leave the uncovered list: no second
build is involved, so there is no region-attribution mismatch (#1066, #1079).
The denominator still moves (456 → 453), the moving-denominator effect described
in #1105. Both figures clear --check-source-regions 80
(.github/workflows/ci.yml:233) and the 86 proposed in #1105.

Scope

Lines 35 and 51 of the hook stay uncovered and are out of scope: every
shipped dialog contains a close button, so !focusable?.length has no
browser-reachable case, and both call sites always pass a live triggerRef, so
the previousFocus fallback has none either. Nothing is left for #1125 to track
once this merges.

Coordination

Claims only tests/e2e/focus-trap-indifferent-keys.spec.js (new) and only
region src/components/hooks/useFocusTrap.js:33. Disjoint from #1120, which
claims tests/e2e/dialog-non-dismissing.spec.js and the backdrop onMouseDown
guards at MemberProfile.js:15 / CommunityPeople/index.js:32. No overlay
under tests/e2e/fixtures/ (#1104, #1117, #1124), no
tests/tools/e2e-coverage-report.mjs (#1040, #1070, #1114), no
tests/tools/e2e-data-fixtures.cjs (#1110), no workflow file.

Related Issue

Closes #1125

---

Human review required.

— 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.