MetaCL Experiment Updates

[mvacaporale]

This PR includes a few loosely related changes:

• Refactoring to compute class indices in two steps L
  • 1. loop through the whole dataset creating a mapping of samples to label
  • 2. partition the samples for each label; this defines a training/testing split within meta-training
• Revising the dataloaders are create so they can share the results from step 1.
• Enhanced logging in ConfigureOptimizerParamGroups to make clear the resulting param group.
• New dendrite experiment that benefits from these changes.