Sync thread subscriptions on boot and reconnect, throttled and staggered

The threads view rendered cache-only on startup, the socket replays
nothing, and the only subscriptions.thread.getView sync ran once per
session on first Threads-view open (sync.Once). After any offline gap
— app closed for days, laptop asleep — the thread_subscriptions table
stayed stale for the whole session.

- Fire EnsureSubscriptions on workspace-ready for every workspace
  (resolved via router.ByID, so background Grid workspaces sync too);
  activation stays as a safety-net trigger.
- Replace the sync.Once with threadSubsGate: at most one paginated
  sweep per 30-minute window per workspace, never two concurrently.
- Stagger each workspace's first sync (slot x 15s) so an Enterprise
  Grid boot doesn't fire N ~62-request sweeps in the same second.
- Kick the gated sync from the reconnect/wake catch-up; the gate
  owns the throttling, so the kick is free when a sync is recent.
- Honor HTTP 429 Retry-After in postForm and retry the same getView
  page, bounded to 3 retries, instead of hard-failing the sweep.