Optimize the eigenvalue method time advance

[namurphy]

Some possibilities include:

• Doing operations directly on numpy.ndarray objects rather than astropy.units.Quantity objects, since doing stuff with units can slow things down.

Additionally, it would be helpful to have the NEI time advance put into a separate function that deals directly with numpy.ndarray objects. That function could then be decorated with the appropriate Numba decorator, or alternatively implemented in Cython.

[StanczakDominik]

One of these solutions contains fewer packaging problems. ;)

[namurphy]

The solve involves creation of a diagonal matrix, which is currently done by creating a full matrix of the appropriate size with numpy.zeros and then setting the diagonal elements to the appropriate values. We could probably make use of functionality within scipy.sparse to represent and perform operations with diagonal matrices more efficiently.