how to sample the log likelihoods

[lea-hagen]

As I'm working on #373, I've been puzzling over the question about the proper parameters for sampling the log likelihoods. Right now, we randomly select 500 from those that have have lnP > max(lnP) - 10. A while back, this selection had been weighted by the probabilities themselves to ensure that we don't miss the highest probability regions, but we decided to switch to random to keep from double-counting the probabilities (issue in #121, PR in #200).

The issue I'm coming across now is whether 500 samples with a -10 threshold is enough to well-sample the probability space. My first question is how these numbers were chosen, and my second question is how to test that they're reasonable.