Now that the map-reduce wordcount is working, I thought it might make a good example to include as a simple illustration of spawn-fetch parallelization. I've tried to add enough comments to make it a good educational tool, explaining what each part does.

Currently the major limitation of this implementation is that the "reduce" is single-threaded. I haven't been able to find a way to perform a key-sort-and-separate step on the HashTables returned by the map step, to allow parallelizing reduce, that isn't just as slow as doing the reduce on a single node anyway.