WIP: Improve emcee

This branch has ballooned a little as I was working on it but I think the set of features is:

• apply a minimum ln likelihood when drawing points from the prior (default leaves behaviour unchanged)
• add npool and map properties to the base_sampler
• support multiprocessing for emcee
• add new diagnostic plots, similar to the plots for ptemcee, also a plot showing how the mean and variance of the log posterior over the ensemble evolve
• calculate the burn_in using the mean and variance of the log posterior (see !842 (merged))
• use some more of the emcee moves
• port some of these changes also into kombine and ptemcee
• add periodic boundaries into ptemcee

This should probably be broken down into pieces.

cc @gregory.ashton @bruce.edelman for ptemcee and kombine.