Add vision model support and related configurations in Assistant

- Introduce new fields in AssistantConfig, schemas, and database models to support vision capabilities, including `vision_enabled` and `vision_model_resource_id`.
- Enhance validation logic in routes to ensure proper handling of vision models and their requirements.
- Update the AssistantPage and related frontend components to include options for enabling vision understanding and selecting appropriate vision models.
- Modify database seed scripts to include vision-related data for assistants, ensuring consistent setup.
- Refactor related functions to integrate vision model handling in the audio-visual processing pipeline.

backend/services/config_resolver.py

@@ -58,6 +58,11 @@ async def resolve_runtime_config(
     stt_resource = await _resource_for(session, assistant.id, "ASR")
     tts_resource = await _resource_for(session, assistant.id, "TTS")
     realtime_resource = await _resource_for(session, assistant.id, "Realtime")
+    vision_resource = (
+        await session.get(ModelResource, assistant.vision_model_resource_id)
+        if assistant.vision_model_resource_id
+        else llm_resource
+    )
 
     return AssistantConfig(
         name=assistant.name,
@@ -89,6 +94,17 @@ async def resolve_runtime_config(
         llm_support_image_input=(
             bool(llm_resource.support_image_input) if llm_resource else False
         ),
+        vision_enabled=assistant.vision_enabled,
+        vision_model_resource_id=assistant.vision_model_resource_id,
+        vision_model=str(_value(vision_resource, "modelId", "")),
+        vision_llm_interface_type=(
+            vision_resource.interface_type if vision_resource else "openai-llm"
+        ),
+        vision_llm_values=(vision_resource.values or {}) if vision_resource else {},
+        vision_llm_secrets=(vision_resource.secrets or {}) if vision_resource else {},
+        vision_llm_support_image_input=(
+            bool(vision_resource.support_image_input) if vision_resource else False
+        ),
         stt_values=(stt_resource.values or {}) if stt_resource else {},
         stt_secrets=(stt_resource.secrets or {}) if stt_resource else {},
         tts_values=(tts_resource.values or {}) if tts_resource else {},
@@ -101,6 +117,12 @@ async def resolve_runtime_config(
         # 运行时连接信息(真 key + url):模型资源优先,否则 .env 兜底
         llm_api_key=_secret(llm_resource, "apiKey", config.LLM_API_KEY),
         llm_base_url=str(_value(llm_resource, "apiUrl", config.LLM_BASE_URL)),
+        vision_llm_api_key=_secret(
+            vision_resource, "apiKey", config.LLM_API_KEY
+        ),
+        vision_llm_base_url=str(
+            _value(vision_resource, "apiUrl", config.LLM_BASE_URL)
+        ),
         stt_api_key=_secret(stt_resource, "apiKey", config.STT_API_KEY),
         stt_base_url=str(_value(stt_resource, "apiUrl", config.STT_BASE_URL)),
         tts_api_key=_secret(tts_resource, "apiKey", config.TTS_API_KEY),