Support for multiple file outputs

[halfak]

When running utilities like diffs2persistence, the major bottleneck is in writing to a compressed stream -- not the actual processing work. In order to take advantage of the distributed processing strategy, we should be writing to multiple output files directly from the mappers.

Each mapper gets a path from which a filename can be extracted.

An optional --output-directory parameter should be provided. Multiple output files (one per input path) should then be written to that directory rather than stdout. An additional argument for compression type should be made available: --output-compression. The default should be bz2.

Note that this will also help do multiprocessing within subsequent stages of the pipeline.