Open a realtime transcription WebSocket

Upgrade to WebSocket, then send {"type":"config","language":"en","interimResults":true} as the first text frame, within five seconds. The type field is required: without it the router answers {"type":"error","code":"INVALID_CONFIG"} and closes. interim_results is accepted as an alias. sttOptions.language, when present, overrides language. The router answers {"type":"ready","provider":"..."}, then send 16 kHz mono 16-bit signed little-endian PCM as binary frames with no WAV header, and finish with {"type":"end"}. Transcripts come back as {"type":"transcript","text":"...","isFinal":bool}; partials replace the previous partial rather than appending. Browsers offer speko.realtime.v1 and speko.ephemeral.<credential> as subprotocols.