Add complex number support to linalg.svdvals

[kgryte]

This PR

• adds complex number support to linalg.svdvals.
• updates the allowed input data type from real-valued floating-point to any floating-point data type.
• requires that the output array have a real-valued floating-point data type having the same precision as the input array.
• makes explicit that, for matrix stacks, singular values must be computed for each matrix in the stack.
• depends on Add complex number support to linalg.svd #561.

[rgommers]

LGTM, in it goes. The interesting bits are in the svd specification. Thanks @kgryte