chore(frontend): kill-switch toggle and global-save retry paths have no tests #573

Summary

Test gaps around the Settings global save and the laddering kill-switch toggle. Found in the independent review of #570 (merged). Related: #539, #570.

Gaps (main at f7405b9)

1. frontend/src/ladder.ts:323-350 (wireKillSwitchToggle) sends { laddering_enabled: toggle.checked } on change, but frontend/src/__tests__/ladder.test.ts never fires the toggle's change event and has no laddering_enabled assertion (the only matches are element ids). Dropping or renaming the field would pass the suite.
2. frontend/src/settings.ts:3553-3560: after await api.updateConfig(changedGlobal) the snapshot is advanced for each sent key. No test fails the global PUT and checks that a retry re-sends the field; the existing failure test (settings.test.ts:622) only covers the error toast. A mutant that advances the snapshot before the PUT survives.
3. Same block: the advance calls getFieldValue(id) after the PUT resolves, while changedGlobal was built from earlier values. A field edited while the PUT is in flight is recorded as saved without having been sent (same shape as snapshotAllFields, settings.ts:370). The test for item 2 should cover this case; fixing it means capturing the sent values before the await.

Acceptance

Tests: the toggle change sends exactly { laddering_enabled: <checked> }; a rejected global PUT followed by Save re-sends the field; an edit made during an in-flight PUT stays dirty afterwards.