feat(agents): 增加模型重试中间件并调整递归限制

将递归限制从100增加到300以支持更深的调用链
添加ModelRetryMiddleware到聊天机器人中间件链
优化模型加载逻辑，支持openai和deepseek官方初始化

Author: xerrors

src/agents/chatbot/graph.py

@@ -1,4 +1,5 @@
 from langchain.agents import create_agent
+from langchain.agents.middleware import ModelRetryMiddleware
 
 from src.agents.common import BaseAgent, load_chat_model
 from src.agents.common.mcp import MCP_SERVERS
@@ -53,6 +54,7 @@ async def get_graph(self, **kwargs):
                 inject_attachment_context,  # 附件上下文注入（LangChain 标准中间件）
                 context_based_model,  # 动态模型选择
                 dynamic_tool_middleware,  # 动态工具选择（支持 MCP 工具注册）
+                ModelRetryMiddleware(),  # 模型重试中间件
             ],
             checkpointer=await self._get_checkpointer(),
         )

src/agents/common/base.py

@@ -74,7 +74,7 @@ async def stream_messages(self, messages: list[str], input_context=None, **kwarg
 
         # 从 input_context 中提取 attachments（如果有）
         attachments = (input_context or {}).get("attachments", [])
-        input_config = {"configurable": input_context, "recursion_limit": 100}
+        input_config = {"configurable": input_context, "recursion_limit": 300}
 
         async for msg, metadata in graph.astream(
             {"messages": messages, "attachments": attachments},

src/agents/common/models.py

@@ -1,11 +1,12 @@
 import os
 import traceback
 
-from langchain.chat_models import BaseChatModel
+from langchain.chat_models import BaseChatModel, init_chat_model
 from pydantic import SecretStr
 
 from src import config
 from src.utils import get_docker_safe_url
+from src.utils.logging_config import logger
 
 
 def load_chat_model(fully_specified_name: str, **kwargs) -> BaseChatModel:
@@ -26,7 +27,13 @@ def load_chat_model(fully_specified_name: str, **kwargs) -> BaseChatModel:
 
     base_url = get_docker_safe_url(model_info.base_url)
 
-    if provider in ["deepseek", "dashscope"]:
+    if provider in ["openai", "deepseek"]:
+        model_spec = f"{provider}:{model}"
+        logger.debug(f"[offical] Loading model {model_spec} with kwargs {kwargs}")
+        return init_chat_model(model_spec, **kwargs)
+
+
+    elif provider in ["dashscope"]:
         from langchain_deepseek import ChatDeepSeek
 
         return ChatDeepSeek(