Refactor vision model configuration handling in voice_webrtc.py

- Update vision model authorization logic to rely solely on server-side configuration, removing client offer dependency for vision capabilities.
- Enhance clarity in comments regarding the role of server configurations in enabling vision understanding.

backend/routes/voice_webrtc.py

@@ -87,7 +87,8 @@ async def _handle_offer(websocket, payload, peers):
         await pc.renegotiate(sdp=offer.sdp, type=offer.type, restart_pc=False)
     else:
         cfg = await _resolve_config(offer)  # 解析放在建连前,配置错就别建连
-        vision_enabled = offer.vision_enabled or cfg.vision_enabled
+        # 服务端助手配置是视觉理解的唯一授权来源；客户端 offer 只负责携带媒体轨。
+        vision_enabled = cfg.vision_enabled
         if vision_enabled:
             has_native_vision = (
                 not cfg.vision_model_resource_id and cfg.llm_support_image_input