#9366 fix(mothership): retry a job dispatch whose usage read failed instead of skipping the run

Summary

Fixes the review findings on the v0.9.33 release PR (#9339) that hold up in the scheduled-jobs lane.

• Usage outage: checkAttributedUsageLimits also refuses with reason: 'usage_unavailable' when the payer's usage can't be read. Dispatch closed such an occurrence permanently as skipped_budget ("The usage limit was reached."). It now counts it as a failed dispatch attempt through the existing bounded retry (retryDispatch: the reconciler retries, and the last allowed attempt ends the run as failed), the way workflow admission treats the same reason as retryable.
• Billing refusals: a run skipped for billing (a blocked account, a payer or member limit) records billing's own message as its summary.
• Test teardown: the job lane suite closes open connections before stopping its fake worker, so a test that fails before ending its stream can't hang afterAll.

Type of Change

• Bug fix

Testing

bun run test:integration lane.integration (real Postgres and Redis): 20 pass. The new test (an outage leaves the occurrence queued with one counted failure; a blocked account records billing's message) fails on the previous dispatch code. tsc --noEmit is clean; job unit tests pass.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)