SPEED: Add support for per-chunk/per-future seeds

(from an old note of mine)

Add support for generating seeds per chunk (=future), i addition to per element as now. Although not invariant to the number of workers, and therefore not numerically reproducible if `nbrOfWorkers()` vary, this will still be statistically sound. The advantage is performance - the number of seeds that need to be pre-generated will be significantly lower when there's a large number of elements.

What should the API be, i.e. how to specify that we want it per chunk without introducing more arguments? Would `future.seed = "per chunk"` vs `future.seed = "per element"` (= `TRUE` as now). Then what about `seed = ` for specifying an initial seed (which now gives "per element" seeds)? Hmm...