We have a field "recovery_start_epoch" in deep_learning_config that can probably be retired

[ant0nsc]

recovery_start_epoch is not settable. We use it to determine the list of training epochs - but even that could be dead code. Verify and clean up

AB#4535