feature request: CartesianIndices with step

[johnnychen94]

To write generic multidimensional function, it is pretty natural to code a CartesianIndex for loop with stride in this way:

stride = 2
R = CartesianIndices(A)
Δ = CartesianIndex(ntuple(_->stride, ndims(A)))

for I in first(R):Δ:last(R)
    ...
end

Unfortunately, this isn't supported, so I have to work it around with something like this:

R_s = map(axes(R)) do I
    first(I):stride:last(I)
end
for I in R[R_s...]
    ...
end

which is quite slow...

I'm wondering if it's okay to get first(R):Δ:last(R) work. Is this already discussed somewhere, or simply not implemented yet? I tried to search the repo but failed to find any related information.

If it's okay to add this, my first try would be to add a new field step to the current CartesianIndices:

struct CartesianIndices{N,R<:NTuple{N,AbstractUnitRange{Int}}, S<:NTuple{N, Int}} <: AbstractArray{CartesianIndex{N},N}
    indices::R
    step::S
end

and then work on the getindex