Fix sparse-dense matmul, with transposed dense

[irhum]

Continuation of the pull request in JuliaGPU/CuArrays.jl#728

As previously mentioned, the patch makes sure the correct method is dispatched

# before the patch
densematrix_T = transpose(densematrix)
@which mul!(similar(densematrix_T, Float32, (size(sparsecscmatrix,1), size(densematrix_T,2))), sparsecscmatrix, densematrix_T)
# mul!(C, A, B) at stdlib/v1.4/LinearAlgebra/src/matmul.jl:208

# after the proposed patch
@which mul!(similar(densematrix_T, Float32, (size(sparsecscmatrix,1), size(densematrix_T,2))), sparsecscmatrix, densematrix_T)
# mul!(C::CuArray{T,2,P} where P, A::Union{CuArrays.CUSPARSE.CuSparseMatrixBSR{T}, CuArrays.CUSPARSE.CuSparseMatrixCSC{T}, CuArrays.CUSPARSE.CuSparseMatrixCSR{T}, CuArrays.CUSPARSE.CuSparseMatrixHYB{T}}, transB::Transpose{#s243,#s242} where #s242<:(CuArray{T,2,P} where P) where #s243) where T at CuArrays/l0gXB/src/sparse/interfaces.jl:20

and tests have been added for the sparse-transpose(dense) case as well

[maleadt]

Ah yes, I really shouldn't have added that typevar. Let's test:
bors r+

[bors]

Build succeeded:

• ci/gitlab/gitlab.com