Persist job logs instead of passing to queue

[jelovirt]

Instead of passing the log to the queue on submit, the worker log should persisted just like output resource. Logs can be very large at debug level, thus they cannot be returned as part of the submission JSON.

At task level, this could be

{
  "output": "s3://bucket/output.zip",
  "log": "s3://bucket/output.log",
}