Draining can kill running jobs

When draining, there's a good chance the runner picks up a job and then the drain kills it.

This race is because the job pick and the job state on Github might have a delay, it might be more reliable to have the SSH command verify if the runner has picked something up.

Since these are ephemeral, a `job picked` or another similar text can be used to find out if a job was picked, in which case the drain call needs to be re-checked by the manager.

Start by tracing the drain call through the manager and the SSH command handling. Check how the runner reports that a job was picked up and how the manager decides draining is complete. Done means a job picked during the race is detected and the drain operation is checked again instead of killing the running job.

Written by the indexing model from the issue text.