Experimentally Drive Feature Enhancements

[Purg]

What does this PR do?

• Adds hyperparameter to select which index of the window to use for prediction and scoring to be something other than just the last frame in the window.
• Added hyperparameter to DropoutFrameDataTransform to fix the drop-out pattern utilized so that it may still be utilized to simulate real-time system performance imperfections but still be highly consistent with it so that test passes may be apples-to-apples comparable. Locs&Confs experiment configurations have been updated to use this parameter for val/test dataset augmentation.
• Sets up the validation DataLoader to also use a weighted random sampler to iterate 3x the size of the validation set in order to get a more class-balanced representation in the validation metrics and plotting, in a similar manner to how training is getting to see class balancing.