Add better support for replicates using silhouette score permutation test

[mparker2]

Introduces a new method for significance testing that does not require pooling of replicates:

• Replicates within and between conditions are compared with EMD to create distance matrix
• Distance matrix and control/treatment labels are used to assess clustering using silhouette
• Reads are permuted between replicates and conditions, and distance matrix/silhouette recomputed
• Null distribution of silhouette scores is fitted with skewed normal distribution, p value estimated.

todo: testing, documentation!