add clustering function for torsions

[cwalker7]

We should add a clustering function which takes in a torsion matrix of [n_frames x n_torsions], using the individual backbone torsions as separate features, and finds the medoid structures. This will be useful for differentiating right-handed and left-handed helices, which the rmsd-based clustering seems to struggle with.

The complication with using torsions as features is that we need to apply periodic boundaries in n-dimensions to compute the distance matrix. The solution here which works for 2 dimensions will need to be modified:

https://francescoturci.net/2016/03/16/clustering-and-periodic-boundaries/

@mrshirts @tlfobe Any thoughts on this?

[cwalker7]

I think I figured out the distance calculation for n-dimensional torsion space - see here: #20