Should the Hermitian requirement of isposdef() be changed to approximately Hermitian?

[ericproffitt]

I'm not sure there's a solution to this question that will make everyone happy. But it seems like for isposdef, where it's implicitly understood that the user likely knows their matrix is Hermitian, and where they're mostly interested in knowing whether all eigenvalues are positive, that it might be worth it to relax the Hermitian constraint to be approximate using isapprox.

What does everyone think about this?

[andreasnoack]

Dup of #182