Training on energy labels

[jonpvandermause]

GPs can now be trained on both force and energy labels. The following modules have been modified to allow this:

• gp.py: the update_db method now takes an optional energy argument.
• gp_algebra.py: the covariance matrix and kernel vectors now include force/energy and energy/energy kernels.
• mgp_en.py: now includes kernels with respect to energy labels, allowing GPs trained on forces and energies to be mapped.

Some notes:

• mgp.py has not been updated; it will only work if the underlying GP was trained on force labels only.
• hyperparameter optimization considers only the force/force block of the covariance matrix. This can be generalized by implementing gradients of the force/energy and energy/energy kernels with respect to the hyperparameters (currently only gradients of the force/force kernel are implemented)