Developments for autoencoders and assess script

[sam-may]

This PR addresses most of the remaining items on the to-do list in #18

Summary:

• implement the option for an AutoEncoder to either (1) train a single autoencoder for all of its histograms or (2) train 1 autoencoder for each histogram. The latter is the new default option.
• fix bug that Kaitlin and Vivan encountered during AutoEncoder.predict() when training with labeled (i.e. good vs. bad) runs
• make AnomalyDetectionAlgorithm classes configurable through json inputs. In particular, this allows for a much smoother method of managing hyperparameters and training options for AutoEncoders
• add SSE histograms on a per-events basis and a per-algorithm basis
• add SSE histograms split by good/bad runs
• add tools to calculate ROC curves, AUC, and their statistical uncertainties
• add tools to plot ROC curves and print efficiency values to tables

Usage of new features is documented in the tutorial