some more debugging to get power law mass1 distributions to work

there was a bad conditional in the definition of NPiecePowerLawMass1 that caused the probabilities to be zero (discovered with PowerLawMass1). There was also a subtle issue with how NPiecePowerLawMass1 looked up the final "mmax" (discovered with BandpassPowerLawMass1) that was confused by the fact that child classes append more parameters beyond the "powerlaw pieces". This also caused the probabilities to be zero.

I also caught and fixed a few issues with check_variates, which we end up calling a lot during each likelihood evaluation. We may want to try to avoid so many repeated calls in the future...