Fix CI tests

[stevengj]

This should fix the failing CI tests — the problem is that, depending on how matrix-multiplication is implemented, A*A may not be exactly Hermitian for a skew-Hermitian A (due to roundoff errors, if different parts of the matrix are computed differently).

[smataigne]

This should fix the failing CI tests — the problem is that, depending on how matrix-multiplication is implemented, A*A may not be exactly Hermitian for a skew-Hermitian A (due to roundoff errors, if different parts of the matrix are computed differently).

Thank you very much for the different updates, I will look at the Schur decomposition. I've already done Cholesky-like but some permutation problems must still be fixed.