[MISC] Clean up torch_npu (#688)

torch_npu 2.5.1 support autoload now. This patch does:
1. remove useless torch_npu import
2. replace `torch_npu.npu` to `torch.npu`.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

tests/ops/test_rotary_embedding.py

@@ -9,7 +9,6 @@ from typing import Optional, Tuple, Union
 import pytest
 import torch
 import torch.nn as nn
-import torch_npu  # noqa: F401
 
 import vllm_ascend.platform  # noqa: F401