feat: Voice Mode - Backend Implementation

[ellaaicare]

Summary

Implement real-time voice conversations with Ella AI. Backend serves as Voice Mode Orchestrator.

PRD

See: backend/docs/PRD_VOICE_MODE_BACKEND_RESPONSE.md

Scope

Component	Effort
Voice mode state machine	4h
SSE client (n8n streaming)	2h
Streaming TTS (OpenAI, modular)	4h
WebSocket binary audio streaming	2h
Voice mode events	2h
/v1/ella/voice-call endpoint	2h

Total: ~2-3 days

Dependencies

• n8n: /webhook/voice-stream SSE endpoint
• iOS: Voice mode UI implementation