v5.10.1

T2 daemon reliability fixes from the 5.10.0 shakeout.

Fixed

• T2 daemon no longer pegs a core on database is locked after a restart
  (nexus-we61e). aspect_queue.reclaim_stale is a global janitor op but ran
  inside every per-process aspect worker's poll loop, so N nx-mcp processes
  RPC'd N redundant reclaim UPDATEs into the single daemon — WAL-lock
  contention that pegged a core after a restart with a stale-row backlog.
  Reclaim now runs once, on a daemon-owned periodic loop (singular by
  construction); workers only claim.
• Lease takeover no longer leaves zero daemons (nexus-64w50). A spawn-lock
  loser that found no reachable winner used to quit, orphaning the service when
  the incumbent was mid-exit in the defer-release-to-exit drain window (lock
  held, discovery file already unlinked). run_t2_daemon now retries the spawn
  so the freed lock is re-acquired. Single-writer is preserved — the spawn lock
  is non-blocking, so a retry wins only when the lock is genuinely free.
• stop() socket teardown is now timeout-bounded (nexus-saigj). The
  wait_closed() calls are capped with _GRACEFUL_STOP_TIMEOUT, so a
  connection draining a long in-flight RPC at SIGTERM can no longer extend the
  spawn-lock hold without bound.
• A restarted daemon clears the stale-row backlog immediately
  (nexus-nhqll). The reclaim loop now reclaims before its first sleep instead
  of waiting a full interval.