Connect GMMClusterer output with Polyssifier

[bbradt]

Get the cluster-assignment probabilities for each instance, reassign these features to subject instances, and feed data back into polyssifier.

The centroids and assignments fields need to be assigned as part of the fit method (most likely using the predict_proba function), and we need to use the reassign_to_subjects function in the dFNC class to handle the multidimensional output to assign cluster probabilities as subject-level features.

E.g. we get the following from predict_proba

a TOTAL_INSTANCES x NUM_CLUSTERS probability matrix.

this needs to be converted into a
NUM_SUBJECTS x NUM_FEATURES matrix, either by taking the argmax of the probability (converting to hard assignment), or by just horizontally concatenating the probabilities over the subject windows. Alternatively, we can look into modifying polyssifier so that it will classify the following tensor:

NUM_SUBJECTS x NUM_WINDOWS x NUM_CLUSTERS

See the reassign_to_subjects function in the dFNC class for info on how this is done currently for the hard-clustering in dFNC

[bbradt]

I believe PR #52 resolves this, but I will leave it open just in case we want to try different ways of connecting the assignments to polyssifier.