Transcribe a short dictation clip from the in-app mic. Provider-abstracted (ASCLEPIUS_STT_PROVIDER: standard = Deepgram/Whisper, wispr stub). Audio is EPHEMERAL — held in memory for this request only, never persisted (synthetic prompts, no PHI; TLS in transit). 503 when no provider is configured so the mic button can degrade to typing.

TUTORIAL rather than full access (PRD A §2.1): dictation is a UI convenience over the user's own microphone, not access to any of our data, and an applicant taking the examination is offered the same mic button everyone else is.