Implement StepFun Realtime service and enhance AssistantConfig

- Add new fields to AssistantConfig for realtime interface configuration, including types, values, and secrets.
- Introduce StepFunRealtimeService to handle speech-to-speech processing via WebSocket, integrating STT, LLM, and TTS functionalities.
- Refactor pipeline execution to support a new realtime mode, allowing direct text input processing and immediate responses.
- Update model resource testing to include validation for StepFun Realtime connections.
- Enhance service factory to create realtime services based on configuration settings.
- Modify README documentation to reflect new realtime capabilities and usage instructions.

backend/routes/voice_webrtc.py

@@ -58,7 +58,9 @@ async def voice_signaling(websocket: WebSocket):
     except Exception as e:
         logger.error(f"WebRTC 信令出错: {e}")
     finally:
-        for pc in peers.values():
+        # disconnect() triggers the registered closed callback, which removes
+        # the peer from this dict. Iterate over a snapshot to avoid mutation.
+        for pc in list(peers.values()):
             await pc.disconnect()