Allow for worker retraining and store new models as Girder objects

[zjniu]

To enable human-in-the-loop training for workers with deep learning models:

• Workers must be able to store new models as Girder objects that belongs to each user and are associated with a particular Docker worker image.
• Workers must be able to access these models when given the Girder object ID (or whatever identifier is necessary).
• Create a new UI element that can be added to worker interfaces that will list all available models associated with that worker and belongs to the current user.
• Allow the user to manage (add/delete/backup) models in the file manager.

[arjunrajlab]

@zjniu We also need to specify the interface for how the worker can send a model to Girder (like, filenames, location, etc.).

[arjunrajlab]

@manthey @bruyeret Would be helpful to know what relevant endpoints exist for regex searching and for pulling/sending files to Girder. Also, how those would be best incorporated into the UPennContrast worker API (if needed).

[manthey]

We have an endpoint in the slicer_cli_web plugin called path_match https://github.com/girder/slicer_cli_web/blob/master/slicer_cli_web/docker_resource.py#L354-L392 that picks a file based on some criteria. We aren't using the slicer_cli_plugin, so perhaps the easiest thing would be to copy this endpoint code to the plugin used in Nimbus Image.