[Fix] fix dashboard restored session rows

[nil-err]

背景 / Problem Found

这次修改是排查线上问题时发现的：botmux 重启前有 3 个会话在 dashboard 上表现为 starting，重启后这些会话在 dashboard 里看不到了。

排查后确认，starting 不是持久化的 Session.status，而是 dashboard 根据 active session 缺少 screen/status 派生出来的展示状态。daemon 重启时 restoreActiveSessions() 会把 active session 注册回内存，但此前不会向 dashboard SSE 广播完整 session row。如果 dashboard 在 hydrate/SSE 时序上先收到后续 session.update / close 事件，却本地还没有对应 row，这些 patch 就会被当成 unknown row 忽略，最终表现为恢复/关闭状态在 dashboard 消失。

What Changed

• Added announceSessionRow(ds) to publish a full session.spawned dashboard row from composeRowFromActive(ds).
• Reused that helper from announcePendingRepoSession() so session row announcements share one path.
• Updated restoreActiveSessions() to announce restored normal sessions and restored adopt sessions immediately after they are registered into activeSessions.
• Added optional wake=1 support to POST /api/sessions/:sessionId/resume.
  • Default behavior is unchanged: resume marks a closed session active and waits for the next inbound message to cold-resume the CLI.
  • With wake=1, the route immediately calls forkWorker(ds, '', true) after a successful resume, which is useful for operational recovery when a session must be usable right away.

Impact

• Restored sessions become visible to dashboard subscribers even when dashboard hydrate and daemon restore race with each other.
• Subsequent status/close updates are applied to a known row instead of being dropped as unknown.
• Existing resume callers keep the same behavior unless they explicitly opt into wake=1.

Validation

• corepack pnpm vitest run --project unit test/dashboard-ipc.test.ts test/dashboard-attention-signals.test.ts test/restore-zombie-close.test.ts
• corepack pnpm exec tsc --noEmit
• git diff --check