Zester v0.6.10

Added

• Presence-aware job finalize (UNREACHABLE fast path): at dispatch the master
  records which targets have no live peel-heartbeat entry
  (Job.OfflineAtDispatch, additive). The hint NEVER gates delivery — every
  resolved target is still published to, and any ack or return overrides it
  (the per-job ack is the authoritative delivery proof) — it only bounds the
  wait: a grace period (default 5s) after the ack-window re-publish, targets
  that were heartbeat-absent AND never acked AND never returned are finalized
  with an explicit synthetic per-peel return (UNREACHABLE: no heartbeat at dispatch and no ack after republish, Return.Unreachable, additive),
  published on the return subject so the dispatching CLI finishes early. A
  ping at a stopped peel now answers in ~10s with an explicit status instead
  of burning the full 60s timeout. Recovery watchers keep deadline semantics.
• Classified CLI exit codes for module executions (job mode and --direct):
  0 all targets succeeded, 2 one or more peels returned a failed result,
  3 one or more peels unreachable/missing (incl. --direct no-responders),
  4 both classes at once; 1 stays the generic CLI/infrastructure error.
  Previously a job-mode timeout with missing returns exited 0. Text output
  gains explicit <peel>: UNREACHABLE status lines, JSON/YAML records carry
  a per-target status field (success/failed/unreachable), and
  zester job show gains a STATUS column in its Returns table.