Make MaxSens partition an iterator

[tomerarnon]

It would be twice as efficient not to have to allocate the partitioned hyperrectangles. Something like

function Base.iterate(P::HyperrectanglePartition, i = 1)
    # Or whatever variables make the most sense:
    low, high = P.low, P.high
    N, Δ, CI = P.N, P.Δ, P.CI

    i > N && return nothing

    I = collect(Tuple(CI[i])) .- 1
    local_lower = @. min(low + Δ*I,       high)
    local_upper = @. min(local_lower + Δ, high)
    H = Hyperrectangle(low = local_lower, high = local_upper)

    return H, i+1
end

Then we allocate only the end result of map(forward_network, iter).

Other potential names: DividedHyperrectangle, SubHyperrectangles.