feat: persistent chat message storage + Android pull-to-load history

Backend:
- Add ChatMessage model + Alembic migration 024
- Add on_message callback to AgentRuntime for persisting messages during SSE streaming
- Plumb session_id from ChatRequest to AgentContext in all 4 chat endpoints
- Add GET /agent-chat/{id}/sessions and /sessions/{sid}/messages with cursor pagination

Android:
- Add DTOs/ApiService/MessageDao for server-side chat history
- ChatRepository: fetchOlderMessages (API + Room cache), offline fallback
- ChatViewModel: loadMoreHistory with isLoadingMore/hasMoreMessages state
- ChatScreen: scroll-to-top detection + top loading indicator

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/app/api/agent_chat.py

@@ -28,9 +28,11 @@ from app.agent_runtime import (
     AgentBudgetConfig,
     AgentMemoryConfig,
     AgentStep,
+    AgentContext,
     AgentOrchestrator,
     OrchestratorAgentConfig,
 )
+from app.models.chat_message import ChatMessage
 from app.core.config import settings
 
 logger = logging.getLogger(__name__)
@@ -68,6 +70,32 @@ def _make_llm_logger(
     return _log
 
 
+def _make_message_saver(
+    db: Session,
+    agent_id: Optional[str] = None,
+    user_id: Optional[str] = None,
+):
+    """创建消息持久化回调，将每条消息写入 chat_messages 表。"""
+    def _save(msg: dict):
+        try:
+            record = ChatMessage(
+                session_id=msg.get("session_id"),
+                agent_id=agent_id,
+                user_id=user_id,
+                role=msg.get("role", "user"),
+                content=msg.get("content"),
+                tool_name=msg.get("tool_name"),
+                tool_input=msg.get("tool_input"),
+                tool_output=msg.get("tool_output"),
+                iteration=msg.get("iteration", 0),
+            )
+            db.add(record)
+            db.commit()
+        except Exception as e:
+            logger.warning("写入 ChatMessage 失败: %s", e)
+    return _save
+
+
 async def _sse_stream(gen: AsyncGenerator[dict, None]) -> AsyncGenerator[str, None]:
     """将 run_stream 生成的 dict 事件格式化为 SSE 文本流。"""
     async for event in gen:
@@ -99,6 +127,43 @@ class ChatResponse(BaseModel):
     token_usage: Optional[Dict[str, Any]] = Field(default=None, description="Token 预算摘要")
 
 
+class MessageItem(BaseModel):
+    """消息历史条目"""
+    id: str
+    session_id: str
+    agent_id: Optional[str] = None
+    user_id: Optional[str] = None
+    role: str
+    content: Optional[str] = None
+    tool_name: Optional[str] = None
+    tool_input: Optional[str] = None
+    tool_output: Optional[str] = None
+    iteration: int = 0
+    created_at: Optional[str] = None
+
+
+class MessageHistoryResponse(BaseModel):
+    """消息历史分页响应"""
+    messages: List[MessageItem]
+    has_more: bool
+    total: int
+
+
+class SessionItem(BaseModel):
+    """会话列表条目"""
+    session_id: str
+    title: Optional[str] = None
+    last_message: Optional[str] = None
+    message_count: int = 0
+    created_at: Optional[str] = None
+    updated_at: Optional[str] = None
+
+
+class SessionListResponse(BaseModel):
+    """会话列表响应"""
+    sessions: List[SessionItem]
+
+
 class OrchestrateAgentItem(BaseModel):
     """编排中单个 Agent 的定义"""
     id: str
@@ -270,7 +335,9 @@ async def chat_bare(
     if req.system_prompt_override:
         config.system_prompt = req.system_prompt_override
     on_llm_call = _make_llm_logger(db, agent_id=None, user_id=current_user.id)
-    runtime = AgentRuntime(config=config, on_llm_call=on_llm_call, streamlined=req.streamlined)
+    on_message = _make_message_saver(db, agent_id=None, user_id=current_user.id)
+    context = AgentContext(session_id=req.session_id)
+    runtime = AgentRuntime(config=config, context=context, on_llm_call=on_llm_call, on_message=on_message, streamlined=req.streamlined)
     result = await runtime.run(req.message)
 
     # 流式美化：为 steps 生成累计摘要
@@ -333,7 +400,9 @@ async def chat_bare_stream(
     if req.system_prompt_override:
         config.system_prompt = req.system_prompt_override
     on_llm_call = _make_llm_logger(db, agent_id=None, user_id=current_user.id)
-    runtime = AgentRuntime(config=config, on_llm_call=on_llm_call, streamlined=req.streamlined)
+    on_message = _make_message_saver(db, agent_id=None, user_id=current_user.id)
+    context = AgentContext(session_id=req.session_id)
+    runtime = AgentRuntime(config=config, context=context, on_llm_call=on_llm_call, on_message=on_message, streamlined=req.streamlined)
     return StreamingResponse(
         _sse_stream(runtime.run_stream(req.message)),
         media_type="text/event-stream",
@@ -417,7 +486,9 @@ async def chat_with_agent(
         config.system_prompt = req.system_prompt_override
 
     on_llm_call = _make_llm_logger(db, agent_id=agent_id, user_id=current_user.id)
-    runtime = AgentRuntime(config=config, on_llm_call=on_llm_call, streamlined=req.streamlined)
+    on_message = _make_message_saver(db, agent_id=agent_id, user_id=current_user.id)
+    context = AgentContext(session_id=req.session_id)
+    runtime = AgentRuntime(config=config, context=context, on_llm_call=on_llm_call, on_message=on_message, streamlined=req.streamlined)
     result = await runtime.run(req.message)
 
     # 流式美化：为 steps 生成累计摘要
@@ -511,7 +582,9 @@ async def chat_with_agent_stream(
         config.system_prompt = req.system_prompt_override
 
     on_llm_call = _make_llm_logger(db, agent_id=agent_id, user_id=current_user.id)
-    runtime = AgentRuntime(config=config, on_llm_call=on_llm_call, streamlined=req.streamlined)
+    on_message = _make_message_saver(db, agent_id=agent_id, user_id=current_user.id)
+    context = AgentContext(session_id=req.session_id)
+    runtime = AgentRuntime(config=config, context=context, on_llm_call=on_llm_call, on_message=on_message, streamlined=req.streamlined)
     return StreamingResponse(
         _sse_stream(runtime.run_stream(req.message)),
         media_type="text/event-stream",
@@ -523,6 +596,127 @@ async def chat_with_agent_stream(
     )
 
 
+@router.get("/{agent_id}/sessions", response_model=SessionListResponse)
+async def list_agent_sessions(
+    agent_id: str,
+    limit: int = 50,
+    current_user: User = Depends(get_current_user),
+    db: Session = Depends(get_db),
+):
+    """获取 Agent 的会话列表，按最近活跃时间排序。"""
+    from sqlalchemy import func as sa_func, desc
+
+    # 验证 agent 存在或有权限
+    agent = db.query(Agent).filter(Agent.id == agent_id).first()
+    if not agent:
+        raise HTTPException(status_code=404, detail="Agent 不存在")
+
+    rows = (
+        db.query(
+            ChatMessage.session_id,
+            sa_func.min(ChatMessage.created_at).label("created_at"),
+            sa_func.max(ChatMessage.created_at).label("updated_at"),
+            sa_func.count(ChatMessage.id).label("message_count"),
+        )
+        .filter(ChatMessage.agent_id == agent_id)
+        .group_by(ChatMessage.session_id)
+        .order_by(desc("updated_at"))
+        .limit(limit)
+        .all()
+    )
+
+    sessions = []
+    for row in rows:
+        # 取第一条 user 消息作为标题
+        first_user_msg = (
+            db.query(ChatMessage)
+            .filter(
+                ChatMessage.session_id == row.session_id,
+                ChatMessage.role == "user",
+            )
+            .order_by(ChatMessage.created_at.asc())
+            .first()
+        )
+        # 取最后一条消息作为预览
+        last_msg = (
+            db.query(ChatMessage)
+            .filter(ChatMessage.session_id == row.session_id)
+            .order_by(ChatMessage.created_at.desc())
+            .first()
+        )
+        sessions.append(SessionItem(
+            session_id=row.session_id,
+            title=first_user_msg.content[:100] if first_user_msg and first_user_msg.content else None,
+            last_message=last_msg.content[:200] if last_msg and last_msg.content else None,
+            message_count=row.message_count,
+            created_at=row.created_at.isoformat() if row.created_at else None,
+            updated_at=row.updated_at.isoformat() if row.updated_at else None,
+        ))
+
+    return SessionListResponse(sessions=sessions)
+
+
+@router.get("/{agent_id}/sessions/{session_id}/messages", response_model=MessageHistoryResponse)
+async def get_session_messages(
+    agent_id: str,
+    session_id: str,
+    before_id: Optional[str] = None,
+    limit: int = 50,
+    current_user: User = Depends(get_current_user),
+    db: Session = Depends(get_db),
+):
+    """获取会话的消息历史（分页），从旧到新排序。"""
+    # limit 限制
+    limit = min(max(limit, 1), 200)
+
+    base_q = db.query(ChatMessage).filter(
+        ChatMessage.agent_id == agent_id,
+        ChatMessage.session_id == session_id,
+    )
+
+    # 游标分页：before_id 之前的老消息
+    if before_id:
+        cursor_msg = db.query(ChatMessage).filter(ChatMessage.id == before_id).first()
+        if cursor_msg and cursor_msg.created_at:
+            base_q = base_q.filter(ChatMessage.created_at < cursor_msg.created_at)
+
+    # 取 N+1 条，判断 has_more（按时间降序取最新 N 条，再反转）
+    batch = (
+        base_q
+        .order_by(ChatMessage.created_at.desc())
+        .limit(limit + 1)
+        .all()
+    )
+
+    has_more = len(batch) > limit
+    if has_more:
+        batch = batch[:limit]
+
+    # 反转为从旧到新
+    batch.reverse()
+
+    total = base_q.count()
+
+    messages = [
+        MessageItem(
+            id=m.id,
+            session_id=m.session_id,
+            agent_id=m.agent_id,
+            user_id=m.user_id,
+            role=m.role,
+            content=m.content,
+            tool_name=m.tool_name,
+            tool_input=m.tool_input,
+            tool_output=m.tool_output,
+            iteration=m.iteration or 0,
+            created_at=m.created_at.isoformat() if m.created_at else None,
+        )
+        for m in batch
+    ]
+
+    return MessageHistoryResponse(messages=messages, has_more=has_more, total=total)
+
+
 def _find_agent_node_config(nodes: list) -> Dict[str, Any]:
     """从工作流节点列表中查找第一个 agent 类型或 llm 类型的节点配置。"""
     if not nodes: