various concatenations involving sparse vectors are dense but should be sparse

[Sacha0]

Ref. #16661/#16722. Concatenations of sparse vectors, combinations of sparse vectors and matrices, and combinations of sparse and dense vectors are consistently sparse. All other concatenations involving sparse vectors fall back to the generic concatenation methods; in such concatenations, the result is often dense when the first argument is sparse, and typically dense when the first argument is not sparse. Some examples:

julia> vcat(rand(4, 1), spzeros(4))
8×1 Array{Float64,2}:
 0.320664
 0.250846
 0.463371
 0.7741
 0.0
 0.0
 0.0
 0.0

julia> hcat(spzeros(4), spzeros(4, 4), rand(4, 2))
4×7 Array{Float64,2}:
 0.0  0.0  0.0  0.0  0.0  0.0623076  0.469735
 0.0  0.0  0.0  0.0  0.0  0.303642   0.210563
 0.0  0.0  0.0  0.0  0.0  0.818678   0.984312
 0.0  0.0  0.0  0.0  0.0  0.71566    0.0121382

julia> hvcat((3,), spzeros(4), rand(4, 2), spzeros(4, 4))
4×7 Array{Float64,2}:
 0.0  0.251215  0.365575  0.0  0.0  0.0  0.0
 0.0  0.413717  0.411106  0.0  0.0  0.0  0.0
 0.0  0.481512  0.363571  0.0  0.0  0.0  0.0
 0.0  0.492115  0.538091  0.0  0.0  0.0  0.0

julia> cat((1,2), rand(4, 2), spzeros(4, 4), spzeros(4))
12×7 Array{Float64,2}:
 0.595454  0.425007  0.0  0.0  0.0  0.0  0.0
 0.155889  0.368775  0.0  0.0  0.0  0.0  0.0
 0.568131  0.926523  0.0  0.0  0.0  0.0  0.0
 0.116905  0.803143  0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0
 0.0       0.0       0.0  0.0  0.0  0.0  0.0

Best!

(Edit: Ref. #17660)