#156 makes the MH sampler handle zero probabilities gracefully. We should make the same changes to the DR and DE samplers.
