Implement precomputed affinity class

[dkobak]

Implements #216.

Allows to do the following:

from openTSNE import TSNE
from openTSNE.affinity import PrecomputedAffinities

# Here P is either a np.array or a scipy.sparse matrix
Z = TSNE().fit(affinities = PrecomputedAffinities(P))

[pavlin-policar]

I think this functionality needs a test. Additionally, would it make sense to re-normalize the affinity matrix here? Or do we want to allow un-normalized affinity matrices? Perhaps a normalize parameter defaulting to True might be the way to go. What do you think?

[dkobak]

Good idea about the normalize parameter. Added it, and added a couple of tests as well.

[pavlin-policar]

Wonderful, thanks! I've added a usage-test, which basically just checks that the whole thing works as intended, then I'll merge.