If slave goes offline while executing a subjob, re-assign subjob

According to the ClusterRunner slave monitor, there are uncommon cases where a build will stay alive permanently (or until we re-deploy ClusterRunner) with one downed slave (red circle) permanently allocated to itself. Consequently, this build never finishes and the client times out.

The correct behavior here would be to de-allocate that slave from this build, and allocate the subjob that was assigned to that slave at the time to a new slave.