Add complex number support to matmul

[kgryte]

This PR

• adds complex number support to matmul.
• requires that input arrays not be transposed or conjugated when arguments have complex floating-point data types. If matmul were to automatically compute the conjugate transpose of the second argument, this would allow returning the complex dot product for the 1D case. However, following NumPy et, this PR pushes conjugate-transposition to userland by requiring conjugation and/or transposition be explicitly performed prior to computing to the matrix product.
• updates the input and output array data types to be any numeric data type, not just real-valued data types.

[rgommers]

That note looks correct; vdot (not in the array API standard), is the matrix multiplication with conjugation. In it goes, thanks @kgryte