fix(UT): rm extra tearDown in test_training.py (deepmodeling#3906)

Fix deepmodeling#3901 .

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

- **Tests**
- Improved test reliability by updating cleanup logic in `test_dp_train`
and `test_trainable` methods.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

source/tests/pt/test_training.py

@@ -66,8 +66,6 @@ def test_dp_train(self):
         )
         trainer_finetune_empty.run()
 
-        self.tearDown()
-
     def test_trainable(self):
         fix_params = deepcopy(self.config)
         fix_params["model"]["descriptor"]["trainable"] = False
@@ -96,8 +94,6 @@ def test_trainable(self):
                 model_dict_before_training[key], model_dict_after_training[key]
             )
 
-        self.tearDown()
-
     def tearDown(self):
         for f in os.listdir("."):
             if f.startswith("model") and f.endswith(".pt"):