FEAT: Add Google Gemini Live API support alongside OpenAI

[Siyarbekir47]

What this PR adds

This extends the original OpenAI-only agent to support Google Gemini Live API as a second provider, selectable at startup — no code changes needed to switch.

Changes

main.py

• Added --openai / --gemini CLI flags (python main.py --openai)
• Refactored into two handlers: _stream_openai() and _stream_gemini()
• Added audio transcoding for Gemini (mulaw 8kHz ↔ PCM 16/24kHz) using audioop
• Fixed double base64 no-op in OpenAI audio forwarding
• Added /incoming-call endpoint for inbound call support
• Added /recording-status endpoint for recording callbacks
• Replaced deprecated ws.open with not ws.closed

requirements.txt

• Added google-genai — Gemini SDK
• Added audioop-lts — audioop replacement for Python 3.13+
• Removed unused openai package (raw WebSocket used instead)

.env.example

• Added GOOGLE_API_KEY, GEMINI_MODEL, GEMINI_VOICE
• Added OPENAI_MODEL, OPENAI_VOICE — now fully configurable
• Added AI_PROVIDER for uvicorn direct usage

README.md

• Full rewrite covering both providers
• Updated voice lists, model options, architecture diagram, all env variables

Usage

python main.py --openai   # OpenAI Realtime API
python main.py --gemini   # Google Gemini Live API