Cluster centroids

[awohns]

Adds code to run louvain community detection and create cluster centroids from the results.
One issue is that clusters with ~ >10000 nodes are difficult to handle, since creating the genotype matrix is extremely expensive (10,000 nodes * 1,000,000 sites). I currently have a max_cluster_size parameter to randomly subsample large clusters.
More generally, we get around the expense of creating huge genotype matrices by using tskit.TreeSequence.simplify(), which cuts down the tree sequence quickly, making it easier to deal with.

[awohns]

On second thought, this might be better as two different functions: one to do the clustering and another to compute the centroids.