A company autoscaling platform should create a clean self-hosted runner for one job and dispose of it afterward. Which runner registration mode best matches that lifecycle?

1. Register one long-lived runner for all teams and delete only its workflow logs.
2. Register an ephemeral runner and arrange its cleanup and external log collection.
3. Use a persistent runner and assume GitHub reimages its operating system after each job.
4. Make every runner repository administrator so it can reset itself between jobs.