separate validation data

[harbecke]

should solve #16 training_data and validation_data are created separately, this lead to the following changes

• hex.training.train.train now takes training_data and validation_data separately
• hex.creation.create_data.create_self_play_data takes num_samples as own argument and from args anymore and has a verbose flag (for validation data)
• torch.save() saves both data in the same file (this can be changed quite easily, if you want)
• config now has train_samples_per_model and val_samples_per_model in [CREATE DATA] instead of samples_per_model and num_validation_data in [TRAIN]

[cleeff]

👍