Fix MultiBrain bug that was introduced with the value estimates (#1018)

python/unitytrainers/ppo/trainer.py

         """
         curr_brain_info = all_brain_info[self.brain_name]
         if len(curr_brain_info.agents) == 0:
-            return [], [], [], None
+            return [], [], [], None, None
 
         feed_dict = {self.model.batch_size: len(curr_brain_info.vector_observations),
                      self.model.sequence_length: 1}

python/unitytrainers/trainer_controller.py

                         (take_action_vector[brain_name],
                          take_action_memories[brain_name],
                          take_action_text[brain_name],
-                         take_action_value,
+                         take_action_value[brain_name],
                          take_action_outputs[brain_name]) = trainer.take_action(curr_info)
                     new_info = self.env.step(vector_action=take_action_vector, memory=take_action_memories,
                                              text_action=take_action_text, value=take_action_value)