Move train parameters to LossFunction

[andrew_peterson]

Many of the parameters in train could properly be moved to LossFunction. The user could then specify these criteria directly through the loss function, something like

calc = Amp( ... )
calc.model.lossfunction = LossFunction(force_coefficient=0.04)

or

calc.model.lossfunction.set('force_coefficient', 0.04)

We should also make this clear in the documentation when we update it.

[andrew_peterson]

We will need to temporarily put in a keyword to train like train_forces=True. Ultimately it should be a little different as we want this to work for other quantities.

[Alireza Khorshidi]

This is fixed in the commit 9d25c95.

Though we need to rethink how we modify force_coefficient based on the convergence values given for energy_rmse, energy_maxresid, force_rmse, and force_maxresid here

[andrew_peterson]

Resolved with issue #60.