Add use_evd option to force EVD over EMI

[scottstanie]

Closes #34

Confirming that use_evd is indeed faster by avoiding the inverse:

In [14]: C_arrs.shape
Out[14]: (200, 200, 25, 25)

In [15]: %timeit mle.mle_stack(C_arrs, use_evd=True, n_workers=16)
882 ms ± 29 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [16]: %timeit mle.mle_stack(C_arrs, use_evd=False, n_workers=16)
2.79 s ± 243 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)