FDR-based Mutect filtering

[davidbenjamin]

In @takutosato's upcoming PR for the new orientation bias tool M2 filtering is going to become a two-pass variant walker in order to filter based on a desired false discovery rate (FDR). We should emulate this for other M2 filters for which is is hard to set a posterior- or likelihood-based threshold. For example, the tumor lod, normal artifact lod, strand bias, contamination, and germline filters all make sense this way.

[davidbenjamin]

And probably the STR / polymerase slippage filter, too.