Large non-hermitian component of plasma response matrix when calculating crit

[jhalpern30]

As discussed in #91, when calculating crit (the smallest eigenvalue of Wp), the original Fortran code forces Hermitian symmetry of the matrix. According to the DCON paper, this matrix should already be Hermitian by construction, so this was confusing. As part of this PR, we added in a warning if the non-Hermitian part of this matrix becomes too large, which I arbitrarily set at 1e-3; however, in some of the basic examples such as the DIIID ideal example, this warning is being raised. It seems like if this warning if being thrown on the simplest of examples, it likely should not be a warning and just be fixed or removed.

I think ultimately, we need to figure out if there being a large non-Hermitian component is a bug that we need to fix and are masking by forcing the Hermitian symmetry or if it not affecting the results and the warning should just be removed.

[logan-nc]

Note that hidden in #24 is an update that cleans up the non-hermitian warnings such that it only prints one waring at the end of the integration telling you how many steps were non-Hermitian (no warning is printed if none were).