Accept prob = difference in posteriors

[JeffreyMinucci]

Although it should have no impact on results, as all priors are uniform, I may want to calculate the posteriors and use their difference instead of the likelihoods to calculate the prob of accepting a worse set of parameters. This should also implicitly eliminate the need to artificial boundary checking, as out of bounds parameters would have prior probability of -Inf.