Close environment if step raises an exception. (#3043)

* close env manager in finally

* rename to env_manager

* remove obsolete mock checks

ml-agents/mlagents/trainers/learn.py

         options.time_scale,
         options.target_frame_rate,
     )
-    env = SubprocessEnvManager(env_factory, engine_config, options.num_envs)
+    env_manager = SubprocessEnvManager(env_factory, engine_config, options.num_envs)
-        curriculum_folder, env, options.lesson
+        curriculum_folder, env_manager, options.lesson
     )
     sampler_manager, resampling_interval = create_sampler_manager(
         options.sampler_file_path, run_seed
     # Signal that environment has been launched.
     process_queue.put(True)
     # Begin training
-    tc.start_learning(env)
+    try:
+        tc.start_learning(env_manager)
+    finally:
+        env_manager.close()
 
 
 def create_sampler_manager(sampler_file_path, run_seed=None):

ml-agents/mlagents/trainers/tests/test_trainer_controller.py

     assert tc.advance.call_count == 11
     tc._export_graph.assert_not_called()
     tc._save_model.assert_not_called()
-    env_mock.close.assert_called_once()
 
 
 @patch.object(tf, "reset_default_graph")
     tf_reset_graph.assert_called_once()
     env_mock.reset.assert_called_once()
     assert tc.advance.call_count == trainer_mock.get_max_steps + 1
-    env_mock.close.assert_called_once()
     tc._save_model.assert_called_once()
 
 

ml-agents/mlagents/trainers/trainer_controller.py

             self._write_training_metrics()
             self._export_graph()
         self._write_timing_tree()
-        env_manager.close()
 
     def end_trainer_episodes(
         self, env: EnvManager, lessons_incremented: Dict[str, bool]