feat: one signal number with remote whisper and in-process fan-out

[daopunk]

Summary

• Collapse Bread Bot to one Signal number (phone B) on the surviving CVM (0e82fa77-8b15-4dbd-89c4-9045ab911353): one signal-api + one signal-bot runs hub, voice, and translation. Drop pairing, PEER_PHONE, and phone A. Do not re-register phone A.
• STT is NEAR AI Whisper Large V3 (GPU TEE). No in-CVM Whisper sidecar. After transcribe, fan spoken text in-process into in-chat auto and Language Threads (this number does not receive its own group posts). Per-message tokio::spawn keeps STT off other handlers.
• Keep live volumes (signal-config-translation, group-prefs-translation, registry-data). In-place phala deploy --cvm-id only.

Test plan

• npm run ci / pnpm run ci (fmt, Clippy, coverage ≥90%, commitlint)
• BOT__ROLE=transcription fail-fasts; translation requires NEAR + WHISPER__ENABLED=true
• One handler stack: Language Threads → voice / !transcribe* → in-chat → menus (!transcription is voice menu only) → quote !translate → !verify / !help / !info / !privacy
• Voice still default off; filenames sent to Whisper are generic (voice.ogg / voice.m4a)
• After STT, in-chat / Language Threads see spoken text as the original speaker; bot translation replies are not re-translated
• !verify is one reply (Bread Bot:); !privacy is one number
• Live (after in-place deploy): phone B still registered, prefs load, menus + !transcribe + translation still work

Follow-ups

• bug: post transcript quote-reply before in-process fan-out #30 — post transcript quote-reply before in-process fan-out
• test: cover speaker-aware transcript fan-out #31 — speaker-aware fan-out tests
• chore: drop leftover two-bot copy and vestigial transcription role wiring #32 — leftover two-bot docs and vestigial role wiring

Out of scope

• Re-registering phone A / proxy :8082
• Migrating group-prefs-transcription
• Dropping BOT__ROLE or putting Whisper back in this CVM