Support for LinearAlgebra.kron

[ytdHuang]

Would it also be possible to support kron for MtlArray (just like CUDA.jl) ?

The current kron for MtlArray uses scalar indexing.

I think we also need the methods for applying kron to Transpose{T,<:MtlArray} and Adjoint{T,<:MtlArray}.

[maleadt]

Should be easy enough to port, as the CUDA.jl implementation only relies on repeat from GPUArrays.

However, if anybody would want to work on this, I'd advise against just copying the code here, but instead moving the existing CUDA kron implementation to GPUArrays.

[ytdHuang]

okay, I re-make a new PR, let's discuss in GPUArrays.jl