update MockClassifierData to support optimal ROC calculations
In essence, we need to provide (analytic?) models of p(features|class) so that measures of the optimal ROC curve could be generated. This boils down to accurately estimating probability distributions over the (high-)dimensional features space generated by the data synthesis algorithm.
This procedure could be as simple as calling MockClassifierData.triggers() to obtain an extremely large number of samples, generating histograms in feature-space, and building a likelihood ratio based on that. However, we may be able to do better by knowing the possible correlations between features that are baked into the data synthesis algorithm.