Propagate flags in train()

[Zachary Ulissi]

It would be convenient to pass in optional keyword arguments to init.train() that would be passed on to the fingerprint calculator and to the model.

For example, it is impossible to pass fortran=True to calculate_fingerprints at the moment.

Alternatively, things could stay the same, and it would be expected that all settings are set at the submodule level. i.e. if fortran=True is set when the fingerprint calculator is made, it is true for all calculations/trainings/etc, but that is not the way the fingerprints and model modules are configured at the moment.

[andrew_peterson]

I think this has been addressed -- please re-open if it still is an issue.