feat(backend): knowledge iteration + vision + neo4j + response contract source

dependencies.py 拆分为 dependencies/ 包; 新增 vision/ragflow_ingestion/neo4j 客户端与 h5_ai_task; alembic 045 图置信度迁移; 响应契约统一收尾。

backend/app/api/conversations.py

@@ -115,9 +115,53 @@ async def list_conversations(
     for emp_id in employee_ids:
         employee_avatar_map[emp_id] = await session_service._get_employee_avatar(emp_id)
 
+    # ── BUGFIX: 批量回退查询员工信息 ──
+    # 为什么需要：conversations 表中 employee_name/department/position 是冗余字段，
+    # 在会话创建时可能为空（异步创建、企微回调延迟等），导致列表API返回空字符串。
+    # 当这些字段为空时，从 employees 表批量查询并回填，确保坐席端能看到完整用户信息。
+    # 何时触发：仅当 conversations 表中的 employee_name 为空字符串时才会去 employees 表查找。
+    employee_name_map: dict[str, dict] = {}
+    empty_name_conv_ids = [
+        conv.employee_id for conv in conversations
+        if not conv.employee_name and conv.employee_id
+    ]
+    if empty_name_conv_ids:
+        try:
+            from app.models.employee import Employee
+            stmt = select(Employee).where(Employee.employee_id.in_(empty_name_conv_ids))
+            result = await db.execute(stmt)
+            for emp in result.scalars().all():
+                employee_name_map[emp.employee_id] = {
+                    "name": emp.name or "",
+                    "department": emp.department or "",
+                    "position": emp.position or "",
+                    "level": getattr(emp, "it_level", "") or "",
+                }
+            if employee_name_map:
+                logger.info(
+                    f"从employees表批量回退获取员工信息: "
+                    f"请求={len(empty_name_conv_ids)}, 命中={len(employee_name_map)}"
+                )
+        except Exception as e:
+            logger.warning(f"从employees表批量回退获取员工信息失败: error={e}")
+
     # 转换为响应 Schema，附加 is_mine / assigned_agent_name / can_grab / avatar 字段
     items = []
     for conv in conversations:
+        # ── BUGFIX: 应用 employees 表回退信息 ──
+        # 如果 conv 的 employee_name 为空，用批量查询结果回填
+        # 这样 ConversationResponse.model_validate 序列化时就能拿到正确的值
+        emp_fallback = employee_name_map.get(conv.employee_id, {})
+        if emp_fallback:
+            if not conv.employee_name and emp_fallback.get("name"):
+                conv.employee_name = emp_fallback["name"]
+            if not conv.department and emp_fallback.get("department"):
+                conv.department = emp_fallback["department"]
+            if not conv.position and emp_fallback.get("position"):
+                conv.position = emp_fallback["position"]
+            if not conv.level and emp_fallback.get("level"):
+                conv.level = emp_fallback["level"]
+
         conv_data = ConversationResponse.model_validate(conv).model_dump()
         # 员工头像（从缓存获取）
         conv_data["avatar"] = employee_avatar_map.get(conv.employee_id, "")
@@ -188,7 +232,7 @@ async def get_conversation(
                 conversation.employee_name = employee.name
                 conversation.department = employee.department or ""
                 conversation.position = employee.position or ""
-                conversation.level = employee.level or ""
+                conversation.level = getattr(employee, "it_level", "") or ""
                 logger.info(
                     f"从employees表回退获取会话详情员工信息: employee_id={conversation.employee_id}, "
                     f"name={employee.name}"
@@ -275,6 +319,7 @@ async def resolve_conversation(
     """结单。
 
     坐席点击"结单"按钮时调用，将会话状态改为 resolved。
+    结单完成后异步触发知识建议生成（通道 A 全链路闭环）。
 
     权限控制：只有主责坐席（assigned_agent_id）才能结单。
     协作坐席和其他坐席不能结单。
@@ -303,6 +348,51 @@ async def resolve_conversation(
     conversation = await session_service.resolve_conversation(conversation_id)
 
     response_data = ConversationResponse.model_validate(conversation).model_dump()
+
+    # ── 任务1（P0）：会话关闭→异步触发知识建议生成 ──
+    # 在结单响应返回后，异步调用 Dify 生成知识迭代建议。
+    # 使用 FastAPI BackgroundTasks 确保不阻塞结单响应。
+    try:
+        from fastapi import BackgroundTasks
+        import asyncio as _asyncio
+
+        async def _trigger_knowledge_suggestion():
+            """异步生成知识建议的后台任务（独立 db session）。"""
+            from app.database import _get_session_factory
+            from app.services.knowledge_iteration_service import KnowledgeIterationService
+
+            factory = _get_session_factory()
+            async with factory() as bg_db:
+                try:
+                    knowledge_service = KnowledgeIterationService()
+                    suggestion = await knowledge_service.generate_knowledge_suggestion(
+                        db=bg_db,
+                        source_type="conversation",
+                        source_data=[str(conversation_id)],
+                        reason=f"会话'{conversation_id}'已结单，自动生成知识迭代建议",
+                    )
+                    if suggestion:
+                        bg_db.add(suggestion)
+                        await bg_db.commit()
+                        logger.info(
+                            f"会话关闭→知识建议已生成: conv_id={conversation_id}, "
+                            f"suggestion_id={suggestion.id}, type={suggestion.suggestion_type}"
+                        )
+                    else:
+                        logger.info(
+                            f"会话关闭→无知识建议生成（Dify不可用或无需建议）: "
+                            f"conv_id={conversation_id}"
+                        )
+                except Exception as e:
+                    logger.error(f"会话关闭→知识建议生成失败: conv_id={conversation_id}, error={e}")
+
+        # 创建后台任务（不阻塞结单响应）
+        _asyncio.ensure_future(_trigger_knowledge_suggestion())
+        logger.info(f"会话结单完成，已触发异步知识建议生成: conv_id={conversation_id}")
+    except Exception as e:
+        # 知识建议生成失败不影响结单主流程
+        logger.warning(f"触发异步知识建议生成失败（不影响结单）: {e}")
+
     return success_response(data=response_data)