Current distributed model storage in ml/validator.py relies on model name.

[mattjhawken]

Current distributed model storage in ml/validator.py relies on model name, which can have duplicates. We must revert to job_id for each model loaded instead as was done with the models_initializing storage.