Stream Recall.ai real-time events (transcript.data, participant events, raw audio) to a webhook or websocket

Documented steps

1. Add recording_config.realtime_endpoints to the POST /bot body; each entry has type ('webhook' or 'websocket'), url, and an events array. These cannot be configured from the dashboard. Ref: https://docs.recall.ai/docs/real-time-webhook-endpoints
2. Webhook form: {"type": "webhook", "url": "https://your-app.com/api/webhook/recall", "events": ["transcript.data", "participant_events.join", "participant_events.leave"]} - the URL must be public HTTPS.
3. Websocket form: {"type": "websocket", "url": "wss://your-app.com/api/ws/audio", "events": ["audio_mixed_raw.data"]} - a continuous stream rather than discrete POSTs. Ref: https://docs.recall.ai/docs/real-time-websocket-endpoints
4. Transcript events: transcript.data for finalized utterances, transcript.partial_data for interim text, transcript.provider_data for the raw provider payload.
5. Participant events: participant_events.join, .leave, .update, .speech_on, .speech_off, .webcam_on, .webcam_off, .screenshare_on, .screenshare_off, .chat_message.
6. Websocket endpoints additionally support raw mixed/separate audio buffers and separate-participant video frames that plain webhooks cannot carry.
7. Authenticate deliveries either with the workspace secret HMAC headers (same `${msgId}.${msgTimestamp}.${payloadStr}` scheme as status webhooks) or by appending ?token=<your-secret> to the endpoint URL. Ref: https://docs.recall.ai/docs/authenticating-requests-from-recallai
8. Real-time webhook retries: up to 60 attempts at a 1-second interval; after that the endpoint is marked failed for the remainder of the call.
9. Real-time websocket retries: up to 30 reconnect attempts at a 3-second interval before the endpoint is marked failed.
10. Set realtime_endpoints at bot creation (or PATCH before the bot joins) - they cannot be added mid-call.