file_reader jobs should be recoverable

[godber]

Teraslice has the ability, when the slicer is implemented correctly, to recover jobs that have been interrupted using the job recovery mechanism:

https://terascope.github.io/teraslice/docs/api#starting-a-job

The file_reader should be able to support this. We should check that, and if not, then make sure it is possible.

[ciorg]

When using recover on a failed job that reads from S3 the job gets stuck in the initializing status. Logs show timeout when connecting to ExecutionController.

The job had 8 failed slices out of 46,000 or so that completed successfully.

Job is using file_assets v2.6.2 and config:

 "_op": "s3_reader",
  "connection": "CONNECTION",
 "path": "PATH/TO/FILE",
 "format": "ldjson",
 "size": 2500000,
 "file_per_slice": false