fix(bin): tmux next-free window index, symlink-safe spawn wait, crew-state lookup retries

[DQ4443]

What Changed

• bin/backends/tmux.sh now creates task windows with the $ses: (trailing colon) target, the documented "next free window index" form, instead of a bare session target that could resolve to the current window and fail with "index in use" under renumber-windows on; this replaces the previous external PATH shim.
• bin/fm-spawn.sh canonicalizes both sides (pwd -P) when waiting for the pane's cwd to move from the project to the treehouse worktree, so a symlinked project path (e.g. macOS /tmp -> /private/tmp) no longer reads as "already moved" and trips the worktree-isolation check.
• bin/fm-crew-state.sh retries the no-mistakes run lookup (configurable FM_CREW_STATE_RETRIES / FM_CREW_STATE_RETRY_DELAY knobs) when a busy CLI returns an empty axi status answer or a timed-out coarse runs listing, instead of surfacing a false stale; answered-but-no-matching-run responses remain authoritative and are not retried. New tests cover the tmux target, the symlinked spawn path, and the retry paths, and the pipeline's review round hardened the timed-out-coarse-lookup case.

Risk Assessment

✅ Low: Well-bounded shell fixes with dedicated regression tests for each; the prior round's retry-gate finding was fixed exactly per the given instructions (NM_RUNS_RESPONDED timeout signature, authoritative no-run fast path preserved) and no new issues surfaced on a full re-pass.

Testing

Completed 1 recorded test check.

Pipeline

Updates from git push no-mistakes

⏭️ **intent** - skipped

✅ No issues found.

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

• ⚠️ docs/scripts.md - merge conflict rebasing onto origin/main

🔧 Fix applied.
✅ Re-checked - no issues remain.

🔧 **Review** - 2 issues found → auto-fixed ✅

• ⚠️ bin/fm-crew-state.sh:394 - The retry loop only retries when the primary axi status call is empty, treating any non-empty answer plus an empty coarse fallback as an authoritative "no run for this branch". But nm_runs_status_for_branch runs no-mistakes runs through the same nm_run timeout wrapper against the same busy CLI, and returns empty both for "list answered, no matching row" (authoritative) and "runs call timed out to empty" (the exact race this loop exists to cover). For a crew whose active run is not the most recent one (concurrent multi-crew validations, the motivating busy-CLI scenario), a responsive axi status reporting another branch's run plus a timed-out runs call breaks without retrying, so the false-stale surfacing can still recur on that path. Fix: distinguish empty-runs-output (retryable, like an empty axi status) from answered-but-no-row (authoritative) and feed that into the cli_responded decision.
• ℹ️ bin/fm-crew-state.sh:363 - Worst-case latency of a crew-state read grows from ~1×FM_CREW_STATE_NM_TIMEOUT (10s) to (1+FM_CREW_STATE_RETRIES)×timeout + retries×delay ≈ 34s at defaults when the CLI is persistently unresponsive. This runs on the watcher's provably-working triage and heartbeat paths, so a hung no-mistakes CLI can stall a poll cycle noticeably longer than before. It is bounded, documented, and configurable (and replaces an equivalent production shim), so no action needed; just noting the tradeoff.

🔧 Fix: retry crew-state lookup on timed-out coarse runs list
✅ Re-checked - no issues remain.

✅ **Test** - passed

✅ No issues found.

• 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.