Check only for Hermitian and not Symmetric in matrix exponentiation (#55349)
For real matrices, a `Hermitian` eigenvalue problem would be equivalent
to a `Symmetric` one, and for complex matrices, there is no special
complex symmetric eigensolver (it errors at present, and complex
symmetric matrices might not be diagonalizable). I think the
`ishermitian` branch should suffice here. Removing the symmetric branch
makes the return type simpler to infer.