fix(cos): start the PR page's Resolve & merge agent immediately

[atomantic]

Summary

Pressing Resolve & merge on an app's Pull Requests tab queued the review-loop follow-up as an auto-approved system task and then left it to the CoS dequeue. That tier (spawnDequeuePriority2AutoApproved) only spawns system tasks while CoS auto-run is in execute and under its daily action budget — so on any other setting the task sat pending until it was started by hand from the task page, which is exactly what the button existed to avoid.

The click is the approval, so /resolve now dispatches the follow-up itself through the same force-spawn path as the task list's Run now button (cos.forceSpawnTask), suppressing the racing dequeue via suppressDequeue — the pattern cosJobRoutes.js already uses for a manually triggered job.

When the spawn genuinely can't proceed (no agent slots, daemon stopped/paused, runner unreachable) the task stays queued in place and the response reports the reason instead of claiming an agent is on it; the tab's toast then says "Started an agent to…" or names what it's waiting for.

Also fixes a related silent gap found while wiring this: a follow-up the task store rejected as a duplicate persisted nothing under the id the service had minted, so callers reported (and would have force-spawned) a task that did not exist. The service now returns the already-queued record with duplicate: true.

Changes

• server/services/agentWorktreeCleanup.js — spawnReviewLoopFollowUp gains dispatch: 'queue' | 'immediate'. queue (default) keeps the autonomous lane exactly as it was; immediate suppresses the dequeue, force-spawns, and reports dispatch: { started, reason }. Duplicate rejections return the existing task.
• server/routes/apps/pullRequests.js — /resolve passes dispatch: 'immediate' and surfaces started + queueReason in the response.
• client/src/components/apps/tabs/PullRequestsTab.jsx — toast/tooltip/help text report what actually happened; a queued-but-not-started result gets the neutral toast with its reason.

Autonomous follow-ups (worktree cleanup, prWatcher escalation) are untouched — the run that produced their PR was itself autonomous, so its continuation stays under the auto-run gate and budget.

Test plan

• server/services/cleanupAgentWorktree.test.js — default dispatch leaves the spawn to the dequeue (suppressDequeue: false, no force-spawn); immediate dispatch suppresses the dequeue and force-spawns the new task id; a refused spawn reports its reason with the task still queued; a duplicate rejection returns the queued record instead of force-spawning a phantom id.
• server/routes/apps/pullRequests.test.js — the route requests an immediate dispatch and reports started: true; a refused dispatch reports started: false + queueReason; a store-rejected duplicate is reported as such.
• client/src/components/apps/tabs/PullRequestsTab.test.jsx — "Started an agent…" on success; the waiting reason surfaced on the neutral toast with the row still showing Queued.
• Full server suite: 38,426 passed / 14 skipped. Full client suite: 10,161 passed / 2 skipped. biome lint clean.

https://claude.ai/code/session_01JErGn4inLCwQiWtXc99m2o