v5.10.5

RDR-151 daemon CPU-peg hardening: a cause-agnostic spin backstop plus the
write-contention root-cause fix.

Fixed

• T2 daemon can no longer spin at ~100% CPU (RDR-151, nexus-u2vmv) — backstop.
  A cause-agnostic event-loop spin guard (spin_guard.py): a SpinGuardSelector
  counts immediate ready-returns by wall-time and a SpinWatchdog daemon thread,
  on a sustained spin, auto-captures the selector map + loop._ready + thread
  stacks and self-heals (SIGTERM → graceful stop; os._exit fallback → clean
  respawn), bounded so a persistent trigger disarms to pegged-but-serving (never
  worse than baseline). Validated against the live postmortem signature.
• T2 daemon write-burst CPU peg root-caused (RDR-151, nexus-xmohw, GH #1137).
  The daemon is now a single serialised writer: every mutating dispatch op AND the
  internal aspect-reclaim loop serialise through one write lock, so a memory.*
  write burst plus the reclaim loop can no longer race the SQLite writer lock into
  a SQLITE_BUSY retry spin. Reads stay concurrent (database.hello is never
  serialised — avoids the slow-hello → ensure-running-stale → takeover cascade);
  lock-hold is bounded to one write attempt (backoff happens outside the lock). A
  coverage forcing-test fails CI if any mutating op is left un-serialised.
• Stop hook no longer files P1 bug beads for transient API failures (#1136).