Not support "tmp[i, k] = A[i,j] * A[j,k]"

[kakulo]

Hey, Looks like TACO does not support "tmp[i, k] = A[i,j] * A[j,k]" on sparse (CSR and COO) on GPU. It only works when using "tmp[i, k] = A[i,j] * transpose({1,0})(k,j)". Could you please check and fix it? Thanks :)