Transcribe speech to text with Whisper (@cf/openai/whisper). POST audio (base64) or audio URL; returns transcript text billed by audio duration.