Add ability to load classifier container with no datasloaders/labels map

[rwood-97]

Describe the bug
MapReader needs you to create a ClassifierContainer in order to run inference on new patches.
Setting this up requires three params:

• model
• dataloaders
• labels_map

The dataloaders and labels map are currently pulled from the annotations so this means that if you are running inference with no annotations then you cannot create the ClassifierContainer.

Would be good to have a separate set up for training vs inference.

Expected behavior
Would like to be able to set up the ClassifierContainer using just the model if all you are doing is inference.