Add a chunking option to GreedyScheduler.

[MasonProtter]

Currently when you work with the GreedyScheduler, every item in the collection has to get put! into and take!en from a Channel which has a lot of overhead if individual iterations are fast.

We could get around this by adding an option to make the greedy scheduler operate on chunks of the the collection instead of items in the collection e.g.

tmapreduce(f, op, coll, scheduler=GreedySchedule(nchunks=100))

would be the equivalent of

tmapreduce(op, chunks(coll; n=100, split=:scatter), scheduler=GreedySchedule()) do inds
    mapreduce(f, op, view(coll, inds))
end