Add an option for PCA to emit the similarity matrix data.

[deflaux]

Right now we compute the similarity matrix and use it immediately within Spark via MLlib to compute the principal components.

It would be great if the job could optionally emit the similarity matrix so that it could be consumed by other machine learning implementations. (e.g., via R or Pandas)