fix(watch): preserve declared-pause stale suppression

[raywest]

Intent

Fix a live, reproducible watcher bug in bin/fm-watch.sh: a done-then-declared-paused crewmate (e.g. a green-CI PR sitting idle awaiting the captain's merge decision) refired a bare 'stale: ' wake on EVERY freshly-armed watcher, even though the .stale- hash suppressor was already correctly set to the current pane hash. Root cause: once a stale hash is already classified for a task whose status log's last line is 'paused:', the watcher re-derives the absorb verdict via crew_absorb_class/pause_state_class on every recheck. A stale/misattributed no-mistakes run-step (e.g. an aborted run whose terminal outcome predates the crew's LATER paused: declaration) can report a non-working, non-paused verdict from fm-crew-state.sh; the old code treated ANY such verdict the same as a fresh 'crew has stopped' first-sight event, firing the bare stale wake and also deleting the .paused- marker as a side effect - which guaranteed the identical misfire on every subsequent poll or watcher restart forever, since crew_absorb_class kept re-reading the same stale aborted-run record. Fix: at all three call sites in bin/fm-watch.sh that consume pause_state_class/crew_absorb_class for a crew whose current status-log line is 'paused:', only a definitive 'working' verdict now breaks the paused treatment; any other verdict (including this misattributed non-working/non-paused case) stays in the existing paused/long-recheck-cadence treatment (handle_paused_stale, anchored on PAUSE_RESURFACE_SECS / the status file's own mtime) instead of surfacing immediately. Genuine wedge detection for a plain (non-paused) stale pane is completely untouched - the escalation timer path (wedge_timer_check, STALE_ESCALATE_SECS) still works exactly as before. Deliberately scoped narrow per the task brief: this fixes only the restart-suppression defect in fm-watch.sh; it does NOT attempt the larger redesign discussed in upstream issues #703 ('Ship task awaiting captain merge is indistinguishable from a wedged crewmate') and #694 ('Watcher cannot hold a wait when the sole in-flight task is a correctly-paused pane') - that boundary is intentional and should be preserved in review. Verification: reproduced the exact live bug in an isolated fixture with a fake fm-crew-state.sh (bare stale refiring on every fresh watcher invocation despite the suppressor already matching the pane hash), added a colocated regression test (tests/fm-watch-triage.test.sh: test_nonterminal_stale_paused_survives_misattributed_run_step_across_restarts) that fails against the pre-fix code and passes post-fix, ran it and the full existing fm-watch-triage.test.sh suite (all green, including all pre-existing declared-pause/resurface/wedge tests, which still validate promptly-detected resumption and untouched wedge escalation), and cross-checked the adjacent watcher/daemon/wake-queue/crew-state/pi-extension suites (all green except one confirmed pre-existing flake in fm-watcher-lock.test.sh that reproduces identically against unmodified HEAD, unrelated to this change). bin/fm-lint.sh (shellcheck 0.11.0, the pinned CI version) is clean.

What Changed

• Keep stale panes on declared-pause handling unless the watcher has definitive working evidence, preventing stale/misattributed run-state reads from repeatedly surfacing bare stale wakes.
• Preserve and throttle paused stale rechecks across watcher restarts while leaving working-pane wedge escalation unchanged.
• Add regression coverage for the restart case and document the declared-pause precedence rule.

Risk Assessment

✅ Low: Captain, the follow-up correctly throttles non-working paused rechecks while preserving the intended narrow stale-suppression behavior.

Testing

Completed 1 recorded test check.

• Outcome: ⏭️ skipped across 5 runs (1h57m55s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

🔧 **Review** - 1 issue found → auto-fixed ✅

• ⚠️ bin/fm-watch.sh:973 - For the persistent none verdict this now absorbs, pause_state_class removes .paused-rechecked-*, so every subsequent poll re-runs crew_absorb_class instead of using its paused recheck throttle. That can invoke the bounded no-mistakes state read every FM_POLL (up to 10 seconds each) for the exact misattributed-run condition. Retain/rearm the recheck marker for non-working paused verdicts while preserving the periodic working-state check.

🔧 Fix: Throttle paused stale rechecks
✅ Re-checked - no issues remain.

⏭️ **Test** - skipped

• 🚨 tests failed with exit code 1
• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; exit "$rc"

🔧 Fix: Preserve declared-pause stale suppression
1 error still open:

• 🚨 tests failed with exit code 1
• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; exit "$rc"

🔧 Fix: Validate watcher stale suppression
1 error still open:

• 🚨 tests failed with exit code 1
• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; exit "$rc"

🔧 Fix: Confirm watcher tests pass
1 error still open:

• 🚨 tests failed with exit code 1
• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; exit "$rc"

🔧 Fix: Preserve declared-pause stale suppression
1 error still open:

• 🚨 tests failed with exit code 1
• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; exit "$rc"

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.