Random Forest has occasional disastrous accuracy: never seen that happen in WEKA
* Refactors KNNClassifier to use them * csv handling moved back into base due to a circular dependency * Also adds the datasets used to test CSV handling