Make use of THOR's checkpointing to continue tasks

[moeyensj]

THOR is checkpointed so that it can continue a run if that run is interrupted. We should support this behavior in thorctl so that we can resume runs if something goes wrong. I think the easiest solution would be to check if the task's output directory already exists in the bucket and, if it does, download the outputs stored in the bucket to the temp directory where THOR will run on the worker.

The function call to download existing outputs could be added here:

thorctl/thorcontrol/taskqueue/client.py

Line 382 in 78be0ba

download_task_inputs_to_dir(bucket, task, input_dir)