Save emcee chain during the run

Currently, if an emcee job crashes or is killed the chain is lost. There's a suggested method for writing each iteration to disk as it's generated here. We should definitely do something like this. Preferably with the ability to resume from an old chain.