CalculateGenotypePosteriors shouldn't modify GTs with no data

[ldgauthier]

A user reported that CalculateGenotypePosteriors generated a bunch of "new" variants (https://gatkforums.broadinstitute.org/gatk/discussion/comment/56226). What happened was that no-calls that look like GT:AD:DP:GQ:PL ./.:0,0:0:.:0,0,0 became GT:AD:DP:GQ:PL:PP 0/1:0,0:0:,:30,0,30 or similar when a high AF prior was applied. The expected behavior is for PPs to not be applied and the GT to remain no-call.