重构架构：恢复统一的 llm_call 节点，移除错误的 final_response 节点

backend/app/main_graph/nodes/react_nodes.py

@@ -3,7 +3,6 @@ React 模式节点模块 - 带超时和重试功能
 包含：
 - react_reason_node: 使用 intent.py 进行推理
 - error_handling_node: 错误处理节点
-- final_response_node: 最终回答节点
 - init_state_node: 初始化状态节点
 
 注意：为了兼容 LangGraph 的同步接口，我们保留了同步的 react_reason 调用
@@ -233,98 +232,6 @@ def error_handling_node(state: MainGraphState) -> MainGraphState:
     return state
 
 
-# ========== 3. 最终回答节点 ==========
-
-from langchain_core.runnables.config import RunnableConfig
-from langchain_core.messages import AIMessage
-
-async def final_response_node(state: MainGraphState, config: RunnableConfig) -> MainGraphState:
-    """
-    最终回答节点：调用 LLM 生成最终回答（支持流式输出）
-    """
-    state.current_phase = "finalizing"
-
-    # 如果已经有 final_result 了，直接返回
-    if state.final_result:
-        state.current_phase = "done"
-        return state
-
-    import time
-    start_time = time.time()
-
-    try:
-        # 构建 LLM 调用链
-        from app.agent.prompts import create_system_prompt
-        from app.model_services.chat_services import get_chat_service
-        from app.logger import debug, info
-        
-        llm = get_chat_service()
-        prompt = create_system_prompt(tools=[])
-        chain = prompt | llm
-
-        # 构建上下文
-        memory_context = getattr(state, "memory_context", "暂无用户信息")
-        
-        # 添加 RAG 上下文到消息
-        messages_with_context = list(state.messages)
-        if state.rag_context:
-            # 把 RAG 上下文作为系统消息添加
-            from langchain_core.messages import SystemMessage
-            rag_system_msg = SystemMessage(content=f"以下是检索到的相关信息：\n{state.rag_context}")
-            # 插入到第一个用户消息之前
-            inserted = False
-            for i, msg in enumerate(messages_with_context):
-                if msg.type == "human":
-                    messages_with_context.insert(i, rag_system_msg)
-                    inserted = True
-                    break
-            if not inserted:
-                messages_with_context.insert(0, rag_system_msg)
-        
-        # 调用 LLM（流式输出）
-        chunks = []
-        async for chunk in chain.astream(
-            {
-                "messages": messages_with_context,
-                "memory_context": memory_context
-            },
-            config=config
-        ):
-            chunks.append(chunk)
-
-        # 将所有 chunk 合并成最终的 AIMessage
-        if chunks:
-            response = chunks[0]
-            for chunk in chunks[1:]:
-                response = response + chunk
-        else:
-            response = AIMessage(content="")
-        
-        elapsed_time = time.time() - start_time
-        
-        # 更新状态
-        state.messages.append(response)
-        state.final_result = response.content
-        state.success = True
-        state.current_phase = "done"
-        state.end_time = datetime.now().isoformat()
-        state.llm_calls = getattr(state, "llm_calls", 0) + 1
-        
-        info(f"⏱️  [LLM统计] 调用耗时: {elapsed_time:.2f}秒")
-        
-    except Exception as e:
-        from app.logger import error
-        import traceback
-        error(f"❌ [LLM错误] 调用失败: {e}")
-        traceback.print_exc()
-        
-        state.final_result = "抱歉，模型暂时无法响应，请稍后再试。"
-        state.success = False
-        state.current_phase = "done"
-
-    return state
-
-
 # ========== 4. 初始化状态节点 ==========
 
 def init_state_node(state: MainGraphState) -> MainGraphState:
@@ -353,11 +260,11 @@ def route_by_reasoning(state: MainGraphState) -> str:
     """
     # 先检查特殊情况
     if state.current_phase == "max_steps_exceeded":
-        return "final_response"
+        return "llm_call"
     if state.current_phase == "error_handling" or state.current_error:
         return "handle_error"
     if state.current_phase == "finalizing" or state.current_phase == "done":
-        return "final_response"
+        return "llm_call"
     if state.current_phase == "retrying":
         if state.retry_action and "rag" in state.retry_action.lower():
             return "rag_retrieve"
@@ -367,7 +274,7 @@ def route_by_reasoning(state: MainGraphState) -> str:
     reasoning_result: Optional[ReasoningResult] = state.debug_info.get("reasoning_result")
 
     if not reasoning_result:
-        return "final_response"
+        return "llm_call"
 
     # 使用 intent.py 提供的路由函数
     route = get_route_by_reasoning(reasoning_result)
@@ -375,18 +282,18 @@ def route_by_reasoning(state: MainGraphState) -> str:
     # 映射到我们的节点名称
     # 注意：这些名称必须与 main_graph_builder.py 中定义的节点名称一致
     route_mapping = {
-        "direct_response": "final_response",
+        "direct_response": "llm_call",
         "retrieve_rag": "rag_retrieve",
         "re_retrieve_rag": "rag_retrieve",
-        "web_search": "web_search",  # ⭐ 新增：联网搜索
-        "clarify": "final_response",
-        "call_tool": "final_response",  # 暂时映射到 final_response，后续可以扩展
+        "web_search": "web_search",
+        "clarify": "llm_call",
+        "call_tool": "llm_call",
         "contact": "contact_subgraph",
         "dictionary": "dictionary_subgraph",
         "news_analysis": "news_analysis_subgraph",
     }
 
-    return route_mapping.get(route, "final_response")
+    return route_mapping.get(route, "llm_call")
 
 
 # ========== 导出 ==========
@@ -394,8 +301,7 @@ def route_by_reasoning(state: MainGraphState) -> str:
 __all__ = [
     "init_state_node",
     "react_reason_node",
-    "web_search_node",  # ⭐ 新增
+    "web_search_node",
     "error_handling_node",
-    "final_response_node",
     "route_by_reasoning"
 ]