[Bug] Runner wait timeout is anchored to queue submission #4341

Summary

The waiting_runner_limit_exceeded timeout appears to be measured from submitted_at. If a job waits in the queue longer than the configured limit, it can be killed shortly after an instance is finally provisioned, before the runner has a reasonable opportunity to check in.

Expected behavior

Start the runner-wait timer at the provisioning transition or another timestamp that represents when the runner could actually start. Audit other timeout calculations for the same queue-time anchoring error.

Regression coverage

A job that queues longer than the runner-wait limit should not fail immediately after provisioning; a genuinely unresponsive runner should still time out after the configured post-provision interval.