`gather` on a sparse source should return a dense array

This is the current behavior:
```julia
julia> s = [1, 2, 3];

julia> t = [2, 3, 1];

julia> A = sparse(s, t, [1.0, 1.0, 1.0], 3, 3)
3×3 SparseMatrixCSC{Float64, Int64} with 3 stored entries:
⋅ 1.0 ⋅
⋅ ⋅ 1.0
1.0 ⋅ ⋅

julia> w = NNlib.gather(A, s, t)
3-element SparseVector{Float64, Int64} with 3 stored entries:
[1] = 1.0
[2] = 1.0
[3] = 1.0
```
I would expect `w` to be a dense vector instead.