joint caller: use phasing information for better calls

[timodonnell]

Once we have the phasing information collected in #389 , besides just writing it out we can also use it to make better calls

Somatic variants should all be "consistent" with germline variants, meaning that reads supporting a somatic variant should either all support the germline variant allele or all support the germline reference allele. Any case where this is not true is a false positive somatic or germline call.

This could be implemented as a filter in #385 , or we could bake it into the likelihood (e.g. give a boost to the prior for variants that overlap a germline variant and are consistent with it, vs. variants that don't overlap a germline variant). [Kind of weird to have a bias toward calling somatic variants in areas near many germline variants, but I guess there's already a lot of bias against calling these regions due to more difficult alignments so maybe it's a good thing.]