v5.6.0

Added

• T2 daemon supervisor & ownership model (RDR-140). Ends the spawn-race /
  SQLite-lock thrash that produced hundreds of t2_daemon_crashed events per day
  when many stacks (CLI sessions, MCP servers, Desktop) started concurrently.
  Four contained/gated phases:
  • Single-flight election. nx daemon t2 ensure-running now takes a
    blocking coordination lock around the discover→spawn decision and
    re-discovers after acquiring it, so K racing stacks converge to exactly one
    cold spawn with the rest attaching — no thundering herd.
  • Loser quiet-attach. A process that loses the spawn lock attaches and
    exits 0 (never opening a second writer) instead of crashing with a
    traceback.
  • Ownership-aware reaping. The startup reap spares a healthy,
    current-version peer (wait-then-force: let a mid-shutdown peer drain, force
    only if it overstays — never coexist) while still reaping stale-version and
    orphaned writers. The RDR-128/129 single-writer backstop is preserved.
  • Crash-loop guard + status. nx daemon t2 status surfaces a
    restarts_in_window count; a bounded guard stops ensure-running respawns
    after repeated failures in a window and logs once, instead of an endless
    crash-loop. (Scope: the ensure-running path; launchd/systemd autostart is
    bounded by the supervisor's own throttle.)
  • Migration cold-start fast-path. When the DB is already at the current
    schema version and in WAL, cold start skips the migration writer-lock via
    lock-free reads.

Changed

• Binary assets are classified as SKIP, not embedded as prose. Textures,
  audio, fonts, and 3D-object files are now skipped at classification time (with
  a skipped non-indexable file log) instead of being embedded as garbage text.