An analog of Hadoop's "Combiner" will reduce the amount of data needed to be written to the global map output for certain workloads. The combiner should have the same interface as `reducer` -- but will need to be run in the mapper before any output is written