Multiprocess optim

[hadware]

Several modifications:

• fixes and PEP8-ization of the API naming convention
• type hinting and fixes to the docstrings
• homogenization of the dissimilarity API
• batch (or chunk)-based processing of dissimilarities for unitary alignments
• leveraging the batch-based processing to compute the dissimilarity for all possible unitary alignments in parallel (using multiprocessing)
• made the batch creation and processing lazy, so the memory isn't affected as much
• fixed the tests to accommodate for all of those changes
• added a documentation stub

Speed ups are noticeable on 4+ multicore CPUs, and on large continuua with 3+ annotators. Statistics are to come.