feat: voice command mapping — context-gated approve/reject over voice

While a tool-permission gate is held on the active tab, spoken confirm
phrases resolve it via the existing approve/reject WS channel — no server
change (byte-shuttle intact). Everything else stays ordinary dictation.

Safety: whole-utterance-exact matching + leading-negation guard + reject
precedence + confidence gate (approve only) + a cancellable 1.5s confirm
window whose commit re-validates gate identity/epoch/connectivity (TOCTOU
guard) + stale-gate epoch bound at PTT-start.

New pure modules public/voice-commands.ts and public/voice-confirm.ts at
100% coverage; wiring in voice/terminal-session/tabs/main. 1307 tests pass,
typecheck clean, build:web OK. Independent code + security reviews flagged a
confirm-window TOCTOU and an unwired cancel path — both fixed and covered.

Plan: docs/PLAN_VOICE_COMMANDS.md.

public/main.ts

@@ -42,8 +42,13 @@ if (joinId) {
 let settings: Settings = loadSettings()
 app.applySettings(settings)
 
-// Key bar (mobile + desktop) sends to whichever tab is active.
-mountKeybar((data) => app.sendToActive(data))
+// Key bar (mobile + desktop) sends to whichever tab is active. onVoiceTrigger
+// activates the full voice path (dictation + context-gated confirm/reject, VC)
+// and un-dormants the 🎤 button (createVoiceInput / matchCommand are otherwise
+// unreachable dead code without this wiring).
+mountKeybar((data) => app.sendToActive(data), {
+  onVoiceTrigger: (action) => app.handleVoiceTrigger(action),
+})
 
 // Multi-device: when this device regains focus, re-assert the active tab's size
 // so a shared session snaps to THIS screen (latest-writer-wins) — full-screen on