make mcmc more efficient! and hopefully better!

This modifies the mcmc sampler in parallel mode. Before, we were appending the initial live point to the list of potential samples. Once we've take a certain number of mcmc walks, we randomly sample from this list to choose a proposal point. The issue is that the list mainly contains samples that are guaranteed to be rejected, and the probability of accepting a good point is greatly reduced. This branch removes that issue, making sampling more efficient and probably more effective at sampling the corners of the posterior.