ChargedProtoANNPID - Optimise MVA calls a little by removing one virtual function call

Currently computing the MVA response for a given set of inputs involves a double virtual function call. One for ANNHelper::getOutput (to dispatch between the TMVA compiled and XML Reader implementations) and then for the TMVA compiled instances, a second for the IClassifierReader interface. This MVA removes the first of these, by effectively promoting the IClassifierReader interface to be what is directly called, and then for the XML Reader based implementation effectively duplicating this same interface as used by the primary compiled instances.