Deepgram streaming speech-to-text via WebSocket. Sends audio chunks as binary frames and receives JSON transcription results in real time.