fix(server): pause Codex goals before stopping turns

[Mjones13]

Stopping a Codex turn can leave its persisted native goal active, allowing the scheduler to inject continuation turns immediately after Stop.

Read the thread goal and pause it only when active, before interrupting the child fleet and root turn. Bound the combined goal lookup/update to one second and ignore goal-control errors so Stop still reaches turn interruption. Extend the collab mock and queued-follow-up regression to assert goal pause precedes the active root turn interrupt.

Validation: original commit 54b359252440a17c59bee6dc77bbf9eaaeac407c was recovered intact from the contributor fork. Sequential source self-review completed; current-upstream synchronization and current-revision CI verification are in progress. Historical reports of 11 focused integration tests plus lint/format/server typecheck are not being presented as fresh results. No merge or deployment is requested by this publication.

Model and harness: GPT-6 Astra Pro, ChatGPT GitHub + Scratch (Level 2), one agent with sequential self-review.

[macroscopeapp]

Approvability

Verdict: Would Approve

Macroscope's review found this PR approvable — This is a narrowly scoped Codex Stop bug fix that pauses an active persisted goal before interrupting turns, with bounded best-effort handling and regression coverage for ordering. The remaining changes extend only the integration-test mock and test assertions.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}