Implement data augmentation to the trainer.play_games_and_train pipeline

[michaelcheah]

When calling trainer.play_games_and_train, no option exists to do augmentation on the generated data sets. This should be done so that the model training can have more data.