Add a mean-log-likelihood method to improve the ACT estimation

Suggestion from @colm.talbot. I've then added a number of other tweaks in addition.

bilby/core/sampler/ptemcee.py

@@ -793,7 +793,7 @@ def check_iteration(
                 "tau usable as {} < gradient_mean_log_posterior={}"
                 .format(gradient_mean_log_posterior, ci.gradient_mean_log_posterior)
             )
-            tau_usable = True
+            tau_usable *= True
         else:
             logger.debug(
                 "tau not usable as {} > gradient_mean_log_posterior={}"