call CNVs even in exome data by finding change points in tumor/normal depth ratio

[timodonnell]

I had mistakenly thought that no one tried calling copy number variants on exome seq due to amplification and capture biases, but apparently the original varscan2 paper found they could get acceptable results if the tumor and normal were prepped in exactly the same way: http://www.ncbi.nlm.nih.gov/pubmed/22300766

The approach is to divide the genome into regions, calculate ratio of tumor / normal depth in each region, and then scan the genome looking for change points in this ratio.

The binning followed by change point detection is similar to the final step for the SV work in https://github.com/hammerlab/guacamole/issues/289 . It may actually be a simpler place to start on that (cc @danvk ).

For validation, the varscan2 paper gives tcga accession IDs where they can compare against copy number arrays in table S1: http://genome.cshlp.org/content/suppl/2012/01/11/gr.129684.111.DC1/Koboldt_SuppTable1.doc