[bug-fix] Fix error with discrete probs (#4309)

ml-agents/mlagents/trainers/torch/utils.py

             entropies = entropies.squeeze(-1)
             all_probs = None
         else:
-            all_probs = torch.cat(all_probs, dim=-1)
+            all_probs = torch.cat(all_probs_list, dim=-1)
         return log_probs, entropies, all_probs