improved definitions of ishermitian/issymmetric

[lxvm]

As mentioned in #1263, the SHermitianCompact type can represent matrices which are neither Hermitian nor symmetric, so this pr fixes the definitions of ishermitian and issymmetric to capture this and adds tests to verify the logic on some of these unusual cases. I think this would be preferable to making SHermitianCompact guarantee that it is in fact Hermitian, i.e. by always returning real values on the diagonal, because the latter would be breaking.

[mateuszbaran]

LGTM. I wouldn't necessarily add a guarantee that SHermitianCompact is hermitian but I think we should fix rand for this type so that it only returns hermitian matrices.

[lxvm]

Thanks, now that the correctness issue of whether the matrix is in fact Hermitian or symmetric is addressed, I don't think we want people to get the impression that this type is in fact mathematically Hermitian by returning random Hermitian matrices. I will add some documentation that explains the caveat that the diagonal may not be real and a formula for making a SHermitianCompact matrix Hermitian, or simply using the Hermitian wrapper.

[mateuszbaran]

OK, let's keep it as-is then.