Build metrics pipeline

[Tennessee-Wallaceh]

• Start with data
• Run classification algorithm
• Store metrics (accuracy, precision, AUC, AUC vs abstention (as laid out in section 5.3 of the paper) )

Should work for variety of classifiers (lay out consistent API).