Added 2nd training workflow: data generator on root files

Created by: hanhiller

Added 'DataGenerator.py', which generates batches of data from root files in the specified (input) directory. Then in 'train.py', three instances (training, valid, and testing) of this class are called to create the feed data, batch by batch, into the keras model and perform the subsequent analysis.

Additionally, when executing 'python train.py', you now specify whether you want train from the root files directly (using the data generator) or convert the root files to h5 and train from these (reading all the data at once). The h5 conversion is now executed within 'train.py.