Parallelization

[bkatiemills]

As discussed in #190-#191, our current parallelization scheme can be much improved by using the multiprocessing infrastructure to distribute individual profiles across a pool of workers as those workers become available, rather than sending fixed batches of profiles to independent processes. The first step to doing this effectively will be to eliminate the dependence of many of the qc tests on global variables which may not be reliable between workers. Some rough ideas on implementation are in branch no-global.