Add docs/voice-dictation-plan.md — a ready-to-implement design covering
server-side Whisper transcription via the existing per-workspace AI provider,
with the mic button in both the AI agent chat and the page editor. The doc
consolidates four parts: STT provider credentials (full parity with the LLM
and embedding creds, incl. the encrypted stt_api_key_enc column and both
provider-field whitelists), the getTranscriptionModel builder + /transcribe
endpoint, the ai.dictation visibility toggle, and the client capture
(useDictation + MicButton). Includes edge cases, security notes, an
implementation order, and the full list of affected files.
