feat(audio): Gemini Live API replaces Whisper+GPT+ElevenLabs

Single WebSocket proxy: frontend PCM16 16kHz → backend → Gemini Live API
Gemini returns PCM16 24kHz audio + text. Playback via Web Audio API queue.
Removed OpenAI/DeepSeek deps. Model: gemini-3.1-flash-live-preview.
Voice: Aoede. Streaming bidirectional audio with silence gating.

backend/config.py

@@ -1,13 +1,8 @@
 from pydantic_settings import BaseSettings
 
 class Settings(BaseSettings):
-    # OpenAI (used for STT + TTS)
-    openai_api_key: str = ""
-
-    # DeepSeek (LLM)
-    deepseek_api_key: str = ""
-    deepseek_base_url: str = "https://api.deepseek.com/v1"
-    deepseek_model: str = "deepseek-chat"
+    # Gemini Live API
+    gemini_api_key: str = ""
 
     # Honcho (memory)
     honcho_api_key: str = ""