fix: 修复编排器使用错误的LLM配置问题 [AC-AISVC-50]

- 移除编排器中硬编码的LLMConfig创建
- 让LLM客户端使用自己的默认配置（从LLMConfigManager获取）
- 修复流式生成方法同样的问题

ai-service/app/services/orchestrator.py

@@ -119,13 +119,7 @@ class OrchestratorService:
             max_evidence_tokens=getattr(settings, "rag_max_evidence_tokens", 2000),
             enable_rag=True,
         )
-        self._llm_config = LLMConfig(
-            model=getattr(settings, "llm_model", "gpt-4o-mini"),
-            max_tokens=getattr(settings, "llm_max_tokens", 2048),
-            temperature=getattr(settings, "llm_temperature", 0.7),
-            timeout_seconds=getattr(settings, "llm_timeout_seconds", 30),
-            max_retries=getattr(settings, "llm_max_retries", 3),
-        )
+        self._llm_config: LLMConfig | None = None
 
     async def generate(
         self,
@@ -345,7 +339,6 @@ class OrchestratorService:
         try:
             ctx.llm_response = await self._llm_client.generate(
                 messages=messages,
-                config=self._llm_config,
             )
             ctx.diagnostics["llm_mode"] = "live"
             ctx.diagnostics["llm_model"] = ctx.llm_response.model
@@ -627,7 +620,7 @@ class OrchestratorService:
         """
         messages = self._build_llm_messages(ctx)
 
-        async for chunk in self._llm_client.stream_generate(messages, self._llm_config):
+        async for chunk in self._llm_client.stream_generate(messages):
             if not state_machine.can_send_message():
                 break