feat: per-forward worker 模型 — 每条转发独立 frpc 配置+独立进程

核心改动（ring 协议不变，Task 本来就是 {Local,Remote,Link} 三元组）:
- process.WorkerKey: worker 键改为 'local~remote~port' 三元组
- renderForward: 每条 forward 渲染只含 1 个 proxy 的独立 frpc 配置
  (替代 renderRemote 把该 remote 全部 forwards 塞进一个进程)
- ClaimFn/RevokeFn: 认领/撤销只操作这一条 forward 自己的进程
- SyncWorkers/auto-start: 只拉起 localOnly 转发的独立进程,
  集群转发由 ring 认领节点拉起 (消除三台争抢 proxy already exists)
- RemoteStatus/StartRemote/StopRemote/RestartRemote: remote 级聚合辅助,
  保持 /profiles API 形状不变, 前端零改动
- handleStatus/logs: 按 worker key 解析真实 remote 名

效果: 三台节点的 frpc 各自只注册自己拥有的 proxy, 单条转发故障不再波及兄弟

internal/process/process.go

@@ -287,6 +287,88 @@ func (m *Manager) ListWorkers() []string {
 	return out
 }
 
+// RemoteWorkers returns the worker keys of all per-forward workers whose key
+// parses to the given remote name. In the per-forward model every link has
+// its own process keyed "local~remote~port"; this aggregates them so remote-
+// centric callers (status page, profile endpoints) can report an overall
+// view without knowing the key format.
+func (m *Manager) RemoteWorkers(remote string) []string {
+	var out []string
+	for _, key := range m.ListWorkers() {
+		if _, r, _, ok := ParseWorkerKey(key); ok && r == remote {
+			out = append(out, key)
+		}
+	}
+	return out
+}
+
+// RemoteStatus summarizes a remote's per-forward workers into one Status:
+// running when ANY forward worker runs; crashed/starting/restarting take
+// priority in that order; stopped (has=false) when no worker exists.
+// This keeps the legacy per-remote status shape working on top of the
+// per-forward worker model.
+func (m *Manager) RemoteStatus(remote string) (Status, bool) {
+	keys := m.RemoteWorkers(remote)
+	if len(keys) == 0 {
+		return Status{State: "stopped"}, false
+	}
+	summary := Status{State: "stopped"}
+	worst := -1 // rank: running=0 < others=1 < crashed=2
+	for _, k := range keys {
+		st, _ := m.Status(k)
+		rank := 1
+		switch st.State {
+		case "running":
+			rank = 0
+		case "crashed":
+			rank = 2
+		}
+		if rank > worst || (rank == worst && summary.State == "stopped") {
+			if rank != worst || st.State != "stopped" {
+				summary = st
+				worst = rank
+			}
+		}
+	}
+	return summary, true
+}
+
+// StartRemote starts every per-forward worker of one remote. Missing workers
+// are skipped; the first error wins.
+func (m *Manager) StartRemote(remote string) error {
+	var firstErr error
+	for _, key := range m.RemoteWorkers(remote) {
+		if err := m.Start(key); err != nil && firstErr == nil {
+			firstErr = err
+		}
+	}
+	return firstErr
+}
+
+// RestartRemote restarts every per-forward worker belonging to one remote
+// (used where the old model had a single per-remote worker). Missing workers
+// are skipped; the first error wins.
+func (m *Manager) RestartRemote(remote string) error {
+	var firstErr error
+	for _, key := range m.RemoteWorkers(remote) {
+		if err := m.Restart(key); err != nil && firstErr == nil {
+			firstErr = err
+		}
+	}
+	return firstErr
+}
+
+// StopRemote stops every per-forward worker of one remote.
+func (m *Manager) StopRemote(remote string) error {
+	var firstErr error
+	for _, key := range m.RemoteWorkers(remote) {
+		if err := m.Stop(key); err != nil && firstErr == nil {
+			firstErr = err
+		}
+	}
+	return firstErr
+}
+
 // StopAll stops all workers (used on manager shutdown).
 func (m *Manager) StopAll() {
 	m.mu.Lock()