Refactor Prediction Service PMML backend example

Different PMML models might return results with a different structure.
As an example, the Logistic Regression and Random Forest models will return respectively (for the Human Task unit test):

approved=1, probability_0=0.10094993308676947, probability_1=0.8990500669132305, predicted_approved=1

and

approved=0.09591028567124724, predicted_approved=0.09591028567124724

The common functionality of the PMML backend (loading models, validating models, evaluating arguments) should be abstracted into its own class.
Users can then extend that class to parse the returned values.