Add multi-processed predictions (and fitting?) to HMMClassifier

[eonu]

• Unlike the KNNClassifier, predictions for the HMMClassifier are still done on a single process, and making predictions can therefore sometimes take a while. Predictions can be multi-processed in a similar way to the KNNClassifier (splitting test data into n chunks, and assigning each chunk to its own subprocess).
• Fitting can also take a while with a large number of GMMHMMs, so it may be worth considering multi-processing for fitting too (divide GMMHMMs into n chunks, and each of the n subprocesses is responsible for fitting the GMMHMMs in its own chunk).

[eonu]

Implemented in #136.