New shuffles setting on HAnDS training data

[harperco]

Currently, FIGER training data for experiment 2 has an alternate mode that generates new random shuffles for incremental increases rather than using the shuffles from the published research.

HAnDS training data should have this same option. Config file already has an entry for it (data["hands"]["newShuffles"]: false) but switching this to true currently has not been implemented.

This would be useful for continuing to test variance and investigate the nature of various good and bad training examples.