Fix crash when next_info is empty and using recurrent

ml-agents/mlagents/trainers/rl_trainer.py

                 agent_brain_info.previous_text_actions[agent_index]
             )
             action_masks.append(agent_brain_info.action_masks[agent_index])
-        if self.policy.use_recurrent:
+        if self.policy.use_recurrent and memories:
             memories = np.vstack(memories)
         curr_info = BrainInfo(
             visual_observations,