Override vcat() and (maybe) promote_rule()

[nalimilan]

We need to override vcat to fix this inconsistency:

julia> vcat(NullableArray(1:2), 1:3)
5-element NullableArrays.NullableArray{Int64,1}:
 1
 2
 1
 2
 3

julia> vcat(1:2, NullableArray(1:2))
4-element Array{Nullable{Int64},1}:
 1
 2
 1
 2

The first case is OK, the second one isn't.

By the way, it could make sense to override promote_rule too, since ideally vcat would use it. Cf. JuliaData/CategoricalArrays.jl#42.

[cjprybol]

issue is with typed_vcat in Base always using the first vcat element to call similar.

julia> similar(full(NullableArray(1:2)), Nullable{Int64}, 4)
4-element NullableArrays.NullableArray{Int64,1}:
 #NULL
 #NULL
 #NULL
 #NULL

julia> similar(full(1:2), Nullable{Int64}, 4)
4-element Array{Nullable{Int64},1}:
 4626872848
 4626872880
 4626872880
 4639814992

See typed_vcat function defintions for AbstractVector & AbstractMatrix. We can fix it by asking typed_vcat to check if any vcat inputs are NullableArrays and, if so, call similar with that input instead of the first.

# fixed
julia> vcat(1:2, NullableArray(1:2))
4-element NullableArrays.NullableArray{Int64,1}:
 1
 2
 1
 2

We need the same change for hcat

# this is wrong
julia> hcat(1:2, NullableArray(1:2))
2×2 Array{Nullable{Int64},2}:
 1  1
 2  2

julia> hcat(NullableArray(1:2), 1:2)
2×2 NullableArrays.NullableArray{Int64,2}:
 1  1
 2  2

And this promote_rule seems wrong? Is promote_rule supposed to be reciprocal?

julia> promote_rule(Nullable{Int64}, Int64)
Nullable{Int64}

julia> promote_rule(Int64, Nullable{Int64})
Union{}

I'll open a pull request with these changes