Add complex number support to linalg.svd

[kgryte]

This PR

• adds complex number support to linalg.svd.
• updates the allowed data types from real-valued floating-point to any floating-point data type.
• requires that the output arrays U and Vh must have the same data type as the input array and that S must be unconditionally real.
• adds an extended description describing the operation.
• makes explicit that, for matrix stacks, the SVD must be computed for each matrix in the stack.

[rgommers]

Additions LGTM, nice to have a bit more detail on what the function does. Thanks @kgryte

[rgommers]

This code comment got removed. I believe that is correct - the comment is wrong. My guess is that it was added when looking at the hermitian boolean keyword for numpy.linalg.svd