fix(go.d): avoid blocking all jobs when stopping a slow job - #21448

The manager previously called job.Stop() while holding the runningJobs mutex.
Since Stop() is a blocking call—it waits until the job’s goroutine exits—this created a scenario where a job stuck in runOnce() could hold the mutex for an extended time.

Summary

Test Plan

Additional Information

Summary by cubic

Moved job Stop() out of the runningJobs mutex to avoid blocking the manager when stopping slow or stuck jobs. This keeps other jobs responsive during stop operations.

• Bug Fixes
  • stopRunningJob now removes the job under lock, unlocks, then calls Stop().
  • startRunningJob calls stopRunningJob before locking, then starts and registers the new job.
  • Prevents global stalls caused by a slow Stop() while holding the mutex.

^{Written for commit eaf7bd4. Summary will update automatically on new commits.}