fix(nodes): 节点健康/connect 绑定 agent 在线状态,不再伪装连上

本次 google 打不开暴露:agent 离线 6 天,节点却一直显示健康、还让客户端
"连上"(凭证推送失败被 _ = 吞掉),实际节点不认 UUID。三处收口:

- Hub.IsOnline(nodeUUID):基于本地 agent 流注册的存活信号(单实例权威;
  多实例需 Redis presence,另议)。
- ListNodes:effectiveNodeStatus 把 DB='up' 但 agent 离线的节点降级为
  "down",不再永远显示健康(status 列只反映供给/调度,不反映 agent 死活)。
- ConnectNode:agent 离线即 503 ErrNodeUnavailable(凭证持久化留待 resync),
  在线时 push 失败也 503——不再吞 pushErr、不再下发节点无法兑现的配置。

测试:TestHub_IsOnline(注册/结束/未注册)、TestEffectiveNodeStatus
(up+离线→down 等)。新增 apierr.ErrNodeUnavailable(503)。

掉线告警 + 静默吞错全量审计另起 TODO。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

server/internal/apierr/apierr.go

@@ -55,6 +55,15 @@ var (
 		MessageZH: "资源不存在",
 		MessageEn: "Resource not found",
 	}
+	// ErrNodeUnavailable (503): a node can't currently serve a connect because its
+	// agent is offline, so a freshly-issued credential can't be provisioned to the
+	// node's sing-box. Returning this instead of a config the node can't honor
+	// stops the client from showing a false "connected" state.
+	ErrNodeUnavailable = &Error{
+		Code:      "NODE_UNAVAILABLE",
+		MessageZH: "节点暂不可用，请稍后重试或换个节点",
+		MessageEn: "Node temporarily unavailable, please retry or pick another node",
+	}
 	ErrConflict = &Error{
 		Code:      "CONFLICT",
 		MessageZH: "资源状态冲突",