fix(openflare): refresh WS read deadline on JSON pong; capture frpc stderr

- read_pump: 收到客户端 JSON {"type":"pong"} 时调用 conn.SetReadDeadline 刷新
  服务端读超时。修复前，服务端仅在 WebSocket 协议层 Pong 帧时刷新 deadline，
  而客户端使用 JSON 应用层 pong 回复，导致服务端 90s 后超时关闭连接，
  客户端收到 EOF 并触发无限重连循环。在 Cloudflare 代理场景下，
  100s 空闲超时进一步加剧了此问题。

- frpc/manager: 捕获 frpc 子进程 stderr 并在进程异常退出时
  将其内容记录到结构化日志 stderr 字段，便于诊断 exit status 1 的
  具体原因（如配置格式错误、Auth Token 失败、relay 服务端不可达等）。

internal/apps/openflare/websocket/read_pump.go

@@ -42,6 +42,13 @@ func runReadPump(
 		case messageTypePing:
 			_ = sendPong(nodeID)
 		case clientPongType:
+			// Refresh read deadline when the client replies with a JSON pong.
+			// This keeps the connection alive when the WebSocket is proxied
+			// through Cloudflare, which enforces a 100-second idle timeout on
+			// the TCP stream. Without this refresh, the server's 90-second read
+			// deadline expires and terminates the connection even though the
+			// client is actively responding to pings.
+			_ = conn.SetReadDeadline(time.Now().Add(wsReadDeadline))
 		default:
 			slog.Debug(logLabel+" unsupported message", "node_id", nodeID, "type", message.Type)
 		}