重构：简化流式架构，将 ReAct 循环移入 agent 节点

主要变更：
- 简化 agent_service：移除复杂双协程，只用 stream_mode=["updates"]
- stream_context：提供更清晰的 API (set_stream_queue/get_stream_queue)
- main_graph_builder：简化图结构，移除 tools 节点和条件边
- agent 节点：包含完整 ReAct 循环 + 流式 Tool Calling 拼接
- 前端：适配新的事件格式
- 添加测试文件：test_full_react_streaming.py, test_stream.py

tools/test/test_full_react_streaming.py

@@ -0,0 +1,104 @@
+"""
+测试新的完整 ReAct 循环架构 + 流式 Tool Calling
+"""
+
+import asyncio
+import sys
+import os
+
+sys.path.insert(0, "/root/projects/ailine/backend")
+
+from app.main_graph.main_graph_builder import build_agent_graph
+from app.model_services import get_cached_chat_services
+from app.agent.stream_context import set_stream_queue
+from app.logger import info, error
+
+
+async def test_full_react_streaming():
+    """测试完整的 ReAct 循环流式架构"""
+    info("=" * 60)
+    info("🧪 测试完整 ReAct 循环 + 流式 Tool Calling")
+    info("=" * 60)
+
+    # 1. 获取服务
+    chat_services = get_cached_chat_services()
+    info(f"✅ 加载了 {len(chat_services)} 个模型: {list(chat_services.keys())}")
+
+    # 2. 构建图
+    graph_builder = build_agent_graph(chat_services, mem0_client=None)
+    graph = graph_builder.compile()
+    info(f"✅ 图构建完成")
+
+    # 3. 创建队列
+    queue = asyncio.Queue()
+    set_stream_queue(queue)
+
+    # 4. 定义后台任务
+    async def run_graph():
+        try:
+            input_state = {
+                "messages": [
+                    {"role": "user", "content": "你好，请介绍一下你自己"}
+                ],
+                "user_id": "test_user",
+            }
+            async for chunk in graph.astream(
+                input_state,
+                stream_mode=["updates"],
+                version="v2"
+            ):
+                await queue.put({
+                    "type": "graph_update",
+                    "data": chunk,
+                })
+        except Exception as e:
+            error(f"❌ 图执行出错: {e}")
+            import traceback
+            error(f"📋 堆栈: {traceback.format_exc()}")
+            await queue.put({"type": "error", "message": str(e)})
+        finally:
+            await queue.put(None)
+
+    # 5. 启动后台任务并处理事件
+    bg_task = asyncio.create_task(run_graph())
+
+    info("\n📡 开始接收流式事件:\n")
+    try:
+        while True:
+            event = await queue.get()
+            if event is None:
+                break
+            if event["type"] == "llm_token":
+                if event["token"]:
+                    print(event["token"], end="")
+                if event["reasoning_token"]:
+                    print(f"<think>{event['reasoning_token']}</think>", end="")
+            elif event["type"] == "turn_start":
+                print(f"\n===== Turn {event['turn']} 开始 =====")
+            elif event["type"] == "tool_start":
+                print(f"\n🔧 工具调用: {event['tool']}")
+            elif event["type"] == "tool_end":
+                print(f"\n✅ 工具调用完成")
+            elif event["type"] == "final_answer":
+                print(f"\n📝 最终答案")
+            elif event["type"] == "graph_update":
+                # 忽略 update 事件，只关心 agent 节点发的事件
+                pass
+            else:
+                print(f"\n📋 其他事件: {event}")
+
+        print("\n✅ 流式测试完成")
+        return True
+
+    except Exception as e:
+        error(f"❌ 测试出错: {e}")
+        import traceback
+        error(f"📋 堆栈: {traceback.format_exc()}")
+        return False
+    finally:
+        if not bg_task.done():
+            bg_task.cancel()
+
+
+if __name__ == "__main__":
+    asyncio.run(test_full_react_streaming())

tools/test/test_stream.py

@@ -0,0 +1,79 @@
+#!/usr/bin/env python3
+"""测试后端流式接口，看看是否真的有流式输出"""
+
+import asyncio
+import aiohttp
+import json
+
+BACKEND_URL = "http://localhost:8079/chat/stream"
+
+
+async def test_stream():
+    print("=" * 60)
+    print("🧪 测试后端流式接口")
+    print("=" * 60)
+
+    async with aiohttp.ClientSession() as session:
+        payload = {
+            "message": "你好，请简单介绍一下自己",
+            "thread_id": "test-thread-001",
+            "model": "zhipu",
+            "user_id": "test-user"
+        }
+
+        print(f"\n📤 发送请求: {json.dumps(payload, ensure_ascii=False)}")
+
+        try:
+            async with session.post(BACKEND_URL, json=payload) as response:
+                print(f"\n✅ 响应状态: {response.status}")
+                print(f"\n📥 开始接收流式响应...\n")
+
+                event_count = 0
+                token_count = 0
+
+                async for line in response.content:
+                    line = line.decode('utf-8').strip()
+                    if line:
+                        if line.startswith("data: "):
+                            data_str = line[6:]
+                            if data_str == "[DONE]":
+                                print("\n🏁 收到 [DONE] 事件")
+                                break
+                            
+                            try:
+                                event = json.loads(data_str)
+                                event_count += 1
+                                print(f"  [{event_count}] {event.get('type')}")
+                                
+                                if event.get('type') == 'llm_token' and 'token' in event:
+                                    token = event['token']
+                                    token_count += 1
+                                    print(f"      → token: {repr(token)}")
+                                
+                                if event.get('type') == 'node_start':
+                                    print(f"      → node: {event.get('node')}")
+                                
+                                if event.get('type') == 'tool_call_start':
+                                    print(f"      → tool: {event.get('tool')}")
+                                
+                                if event.get('type') == 'tool_call_end':
+                                    print(f"      → tool: {event.get('tool')}")
+                                
+                                if event.get('type') == 'error':
+                                    print(f"      ❌ 错误: {event.get('message')}")
+                                
+                            except Exception as e:
+                                print(f"  ❌ 解析失败: {e}, 原始数据: {repr(data_str)}")
+                        else:
+                            print(f"  📝 原始行: {repr(line)}")
+
+                print(f"\n📊 统计: {event_count} 个事件, {token_count} 个 token")
+
+        except Exception as e:
+            print(f"\n❌ 请求异常: {e}")
+            import traceback
+            print(f"📋 堆栈: {traceback.format_exc()}")
+
+
+if __name__ == "__main__":
+    asyncio.run(test_stream())