feat(uq-models): implement deep ensemble and MC dropout models, add UQ visualizations to PostProcessor.

[khakis23]

UQ Models Implementation

This feature adds two new UQ model wrappers, deep ensemble (DE) and MC dropout (MCD), which introduce additional visualizations into the post processor.

New Classes:

• DeepEnsemble
  • (DeepEnsembleHyperModel)
• MCDropout
  • (MCDropoutHyperModel)
• UQVisualizer

Key Features:

• Models integrates with existing model infrastructure (Tuner, PostProcessor, etc.)
• Negative Log Likelihood loss function is added for neural networks
• New UQ visualizer class that integrates into Post Processor
  • New uncertainty visualization method wrapper in Post Processor
  • Integrates variance features into 3 existing Post Processor visualization methods automatically

Changes:

• Check for UQ parameters in Tuner
• Post Processor activates ensemble mode for DE's
• Post Processor owns a UQVisualizer for use in 3 existing and 1 new method
• PostProcessor._models now contains additional "Models" key with the value as the model instance
• Implements NLL in utils.py
  • With validation to ensure the final output layer is the correct size: an error will be thrown if incorrect
• Adds unit test
• Adds integration tests
• Includes usage examples in chf.ipynb, bwr.ipynb, and mit_reactor.ipynb