should reductions return dense arrays?

[CarloLucibello]

Reductions on SparseMatrixCSC return dense arrays. Should the same happen for GBMatrix? Current behavior is

julia> using SparseArrays, SuiteSparseGraphBLAS

# dense matrices from sparse arrays' reductions for base types
julia> sum(sprand(5,5,0.1); dims=1)
1×5 Matrix{Float64}:
 0.0  0.0  0.623837  0.0  0.751621

julia> sum(sprand(5,5,0.1); dims=2)
5×1 Matrix{Float64}:
 0.0
 1.303100775231686
 0.09521454262437534
 0.34834841964204744
 1.4844961238290506

# sparse matrices from sparse arrays' reductions for this repo

julia> sum(GBMatrix(sprand(5,5,0.1)); dims=2)
5x1 GraphBLAS double matrix, bitmap by col
  1 entry, memory: 256 bytes

    (4,1)    0.9055


julia> sum(GBMatrix(sprand(5,5,0.1)); dims=1)
5x1 GraphBLAS double matrix, bitmap by col
  3 entries, memory: 256 bytes

    (2,1)    0.377595
    (3,1)    0.114248
    (5,1)    0.18736