hard-coded number of parallel processes in infer function

[bcollica]

The infer() function accepts an argument for the number of parallel processes to use, but then it hard codes the number of processes in the update xis step to be 2. See below:

xis = _update_xis(sample_features,
                      difference_matrices,
                      difference_penalty,
                      gamma=gamma,
                      n_parallel_processes=2,
                      verbosity=0)

This forces inference to be very slow.

[bcollica]

Fixed in #21