`eigmax` and `eigmin` claim complex numbers cannot be sorted

[araujoms]

The docstrings of eigmax and eigmin write that "Note that if the eigenvalues of A are complex, this method will fail, since complex numbers cannot be sorted.". This is nonsense, the complex numbers are a totally ordered set. I'd like to fix that. There are two solutions:

1. Rewrite the docstring to say that there is no canonical order for the complex numbers.
2. Change the function to start sorting complex numbers (by lexicographical order, for instance).

I think the latter is more useful, but I'm also fine with the former. Opinions?

[ViralBShah]

Cc @stevengj

[stevengj]

I would just change the docstring to clarify it as needed. Note, however, that it's not just a lack of a "canonical" ordering … as if mathematicians have simply not made up their minds yet … it's that there is no ordering consistent with the algebra — they are not an "ordered field".

Of course, we could use the lexicographical ordering for eigmin and eigmax if we want, but I would suggest that:

1. Using eigmin or eigmax on complex eigenvalues is likely to be a bug. If they really want the lexicographical ordering they should request it explicitly.
2. There is no computational advantage to providing these functions for complex eigenvalues versus asking the user to compute all the eigenvalues with eigvals and then picking the one they want. (Unlike the Hermitian case, where we can really compute just the minimum or maximum eigenvalue more efficiently.)