fix(omp): keep supervision outcome delivery responsive

[dnth]

Intent

Port upstream firstmate PR kunchenguid#3767 (fix(pi): keep supervision outcome delivery responsive) into this fork (dnth/firstmate).

The persistent supervision branch delivers wake outcomes to the main session: routine verdicts merge back with no turn and captain verdicts open one follow-up turn. Keep that outcome DELIVERY responsive so main receives supervision outcomes promptly and supervision does not silently fall behind.

This is a bounded PORT-ADAPT, not a clean cherry-pick. The upstream Pi diff was inspected at f09de8a. This fork runs an OMP-adapted supervision branch in bin/fm-supervise-daemon.sh plus .omp/extensions, landed as fork PR #63, not the Pi extension. Adapt the shared bash-layer implications directly where applicable, re-express Pi-extension behavior against the OMP branch, and reconcile with the fork's multibrain delivery path and watcher-continuity fix PR #99 in bin/fm-primary-watch-core.ts. If a Pi piece has no OMP equivalent, record the deliberate omission rather than force-fitting it.

Acceptance criteria:

• AC1: The supervision-outcome-delivery responsiveness fix from fix(pi): keep supervision outcome delivery responsive kunchenguid/firstmate#3767 is present in the fork supervision path, adapted to the OMP supervision branch and not left as a Pi-only change.
• AC2: A colocated regression test demonstrates supervision outcomes deliver back to main responsively with no stall under the fix(pi): keep supervision outcome delivery responsive kunchenguid/firstmate#3767 scenario.
• AC3: Existing supervision/branch tests pass, touched bin scripts are shellcheck-clean, the TS extension is type-clean, and wake routing is not regressed.
• AC4: A short reconciliation note records what fix(pi): keep supervision outcome delivery responsive kunchenguid/firstmate#3767 changed in the Pi extension, how each part maps onto the fork OMP branch, and deliberate adaptations or omissions.

Current implementation decisions: use an awaited OMP async subprocess executor and awaitable OMP grant helpers; serialize OMP ownership checks, outcome append/handoff, and visible report delivery in a delivery queue; retain synchronous ownership checks only for OMP's synchronous offer handshake and shell spawn hook; include the helper in the OMP branch marker hash; leave bin/fm-supervise-daemon.sh process-isolated and preserve bin/fm-primary-watch-core.ts watcher continuity; omit Pi-only processed-outcome reconciliation and Pi renderer/live-TUI guards because OMP has no equivalent path.

Do not merge the PR. Do not push the default branch. Keep the change on the feature branch and deliver a PR against main.

Firstmate-Validation-Generation: e9df3a5a6b498b4627d9afdb25c224dc

What Changed

• Added awaited subprocess execution and a serialized OMP delivery queue so supervision outcome append, cursor handoff, ownership checks, and visible reports stay ordered without blocking the event loop.
• Updated OMP branch helper/version hashing and propagated the async executor through fixtures, live/type checks, primary/secondmate coverage, and a regression for responsive ordered delivery.
• Documented the upstream fix(pi): keep supervision outcome delivery responsive kunchenguid/firstmate#3767 reconciliation, including OMP-specific adaptations, preserved watcher continuity, and deliberate Pi-only omissions.

Risk Assessment

✅ Low: The port is internally consistent: delivery-side subprocesses are awaited and serialized, synchronous checks remain only at required handshake hooks, marker hashing includes the new helper, and the added regression exercises event-loop responsiveness and ordering without source-content-only assertions.

Testing

Ran the focused supervision regression (including the event-loop responsiveness/order scenario), wake routing and primary/session tests, and strict no-emit OMP typecheck successfully; no UI surface exists for screenshot evidence. The real-SDK live guard was also attempted but failed because the broken-branch degrade path did not deliver the expected watcher-wake steer to main.

Evidence: Async delivery regression evidence

ok - OMP outcome delivery yields to the event loop while preserving routine/captain order
TICKS=24
DELIVERIES=routine,captain

Evidence: Live OMP guard failure

not ok - degrade-to-main guard failed: DRIVER_FAIL: a broken branch did not fall the wake back to main via a watcher-wake steer

- Outcome: ⚠️ 1 error across 1 run (8m30s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

⚠️ **Test** - 1 error

• 🚨 tests/fm-omp-branch-live-e2e.test.sh - The real-SDK live guard failed in its degrade-to-main scenario: after an intentionally invalid branch model, no watcher-wake steer reached main within the 120-second wait. Portable regression and targeted routing tests pass, but end-to-end fallback behavior remains unproven and may indicate an OMP SDK/environment incompatibility or a product regression.
• bash tests/fm-omp-branch-supervision.test.sh
• bash tests/fm-omp-primary.test.sh
• bash tests/fm-send-turn-start.test.sh
• bash tests/fm-session-start.test.sh
• bash tests/fm-wake-queue.test.sh
• bash tests/fm-omp-branch-types.test.sh
• FM_OMP_BRANCH_LIVE_E2E=1 bash tests/fm-omp-branch-live-e2e.test.sh

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.