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

@@ -67,7 +67,7 @@ class Ptemcee(MCMCSampler):
     autocorr_tau: int, (1)
         The number of autocorrelation times to use in assessing if the
         autocorrelation time is stable.
-    gradient_tau: float, (0.05)
+    gradient_tau: float, (0.1)
         The maximum (smoothed) local gradient of the ACT estimate to allow.
         This ensures the ACT estimate is stable before finishing sampling.
     Q_tol: float (1.01)
@@ -773,10 +773,10 @@ def print_progress(
 
     sampling_time = datetime.timedelta(seconds=np.sum(time_per_check))
 
-    if grad >= 0:
-        tau_str = "{}(+{:0.2f})".format(tau_int, grad)
+    if grad < convergence_inputs.gradient_tau:
+        tau_str = "{}({:0.2f}<{})".format(tau_int, grad, convergence_inputs.gradient_tau)
     else:
-        tau_str = "{}({:0.2f})".format(tau_int, grad)
+        tau_str = "{}({:0.2f}>{})".format(tau_int, grad, convergence_inputs.gradient_tau)
 
     if tau_usable:
         tau_str = "={}".format(tau_str)