parallelize set difference

[johnyf]

Either using threading to avoid the cost of creating new processes with multiprocessing, or with a persistent pool of workers using multiprocessing, to which ConvexPolytope - ConvexPolytope jobs are passed.

The former approach is arguably preferable, because the computation tree is significantly parallelizable, due to the intermediate production of non-convex polytopes, which enable further differences to be passed to freshly spawned threads.

Note that if parallelization results in a significant difference, then that is expected to be observable also on personal machines, for those cases the branching factor of a difference (the number of convex polytopes into which a convex polytope breaks after subtracting another convex polytope) is not much larger than 4 (a common number of processors on laptops).