Fix sparse matrix adaption

[Abdelrahman912]

Right now, adaption of sparse matrix gives the same CPU matrix:

using AMDGPU, SparseArrays
import KernelAbstractions as KA, Adapt
A = spdiagm(0 => ones(5), -1 => -ones(4), 1 => -ones(4))
typeof(Adapt.adapt(ROCBackend(), A))   # print SparseMatrixCSC, not ROCSparseMatrixCSC

[luraess]

Thanks for the fix.