修复：P0+P1 生产稳定性和性能优化（6项）

P0 稳定性修复：
- 告警去重字典添加惰性清理机制，防止长时间运行内存溢出
- Redis 连接断开时显式 close() 后再置 None，防止文件描述符泄漏
- 截图消息 ACK 移至成功路径，失败消息留在 pending list 自动重试

P1 性能优化：
- GPU NMS 添加 torch.no_grad() + 显式释放临时张量，减少显存碎片
- 截图存储改为 Redis 原始 bytes，去掉 Base64 编解码开销（兼容旧格式）
- ROI 配置查询 N+1 改为 get_all_bindings() 单次 JOIN 查询

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

core/screenshot_handler.py

@@ -59,6 +59,7 @@ class ScreenshotHandler:
 
         self._thread: Optional[threading.Thread] = None
         self._stop_event = threading.Event()
+        self._last_pending_check = 0.0
 
     # ==================== 生命周期 ====================
 
@@ -180,20 +181,26 @@ class ScreenshotHandler:
 
                 backoff = 5  # 重置退避
 
+                # 每 60 秒检查一次 pending 消息
+                import time as _time
+                if _time.time() - self._last_pending_check > 60:
+                    self._last_pending_check = _time.time()
+                    self._cleanup_pending_messages()
+
                 for stream_name, messages in results:
                     for msg_id, fields in messages:
                         try:
                             self._handle_request(fields)
-                        except Exception as e:
-                            logger.error("[截图] 处理请求失败: %s", e)
-                        finally:
-                            # ACK 消息
+                            # 处理成功才 ACK
                             try:
                                 self._cloud_redis.xack(
                                     SNAP_REQUEST_STREAM, SNAP_CONSUMER_GROUP, msg_id
                                 )
                             except Exception:
                                 pass
+                        except Exception as e:
+                            logger.error("[截图] 处理请求失败 (msg_id=%s): %s", msg_id, e)
+                            # 不 ACK，消息留在 pending list 等待重试
 
             except redis.ConnectionError as e:
                 if self._stop_event.is_set():
@@ -409,3 +416,38 @@ class ScreenshotHandler:
             logger.info("[截图] 降级写 Redis 成功: request_id=%s", request_id)
         except Exception as e:
             logger.error("[截图] 降级写 Redis 也失败: %s", e)
+
+    # ==================== Pending 消息清理 ====================
+
+    _MAX_RETRY_COUNT = 3
+    _PENDING_IDLE_MS = 30000  # 消息 pending 超过 30 秒才处理
+
+    def _cleanup_pending_messages(self):
+        """清理 pending list 中重试次数过多的消息"""
+        try:
+            pending = self._cloud_redis.xpending_range(
+                SNAP_REQUEST_STREAM, SNAP_CONSUMER_GROUP,
+                min="-", max="+", count=50,
+                consumername=self._consumer_name
+            )
+            for entry in pending:
+                msg_id = entry['message_id']
+                delivery_count = entry['times_delivered']
+                idle_ms = entry['time_since_delivered']
+
+                if idle_ms < self._PENDING_IDLE_MS:
+                    continue
+
+                if delivery_count > self._MAX_RETRY_COUNT:
+                    logger.warning(
+                        "[截图] 消息超过最大重试次数，丢弃: msg_id=%s, retries=%d",
+                        msg_id, delivery_count
+                    )
+                    try:
+                        self._cloud_redis.xack(
+                            SNAP_REQUEST_STREAM, SNAP_CONSUMER_GROUP, msg_id
+                        )
+                    except Exception:
+                        pass
+        except Exception as e:
+            logger.debug("[截图] 检查 pending list: %s", e)