Right loss function for stability, fix some pypi

ml-agents/mlagents/trainers/coma/optimizer_torch.py

         )
         loss = (
             policy_loss
-            + 0.5 * (value_loss + baseline_loss)
+            + 0.5 * (value_loss + 0.5 * baseline_loss)
             - decay_bet * ModelUtils.masked_mean(entropy, loss_masks)
         )
 
             modules.update(reward_provider.get_modules())
         return modules
 
-    def get_trajectory_value_estimates(
+    def get_trajectory_and_baseline_value_estimates(
-    ) -> Tuple[Dict[str, np.ndarray], Dict[str, float]]:
+    ) -> Tuple[Dict[str, np.ndarray], Dict[str, np.ndarray], Dict[str, float]]:
 
         n_obs = len(self.policy.behavior_spec.observation_specs)
 

ml-agents/mlagents/trainers/coma/trainer.py

             value_estimates,
             baseline_estimates,
             value_next,
-        ) = self.optimizer.get_trajectory_value_estimates(
+        ) = self.optimizer.get_trajectory_and_baseline_value_estimates(
             agent_buffer_trajectory,
             trajectory.next_obs,
             trajectory.next_group_obs,