[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/multicard/test_pyhccl_distributed.py

@@ -20,7 +20,6 @@ import multiprocessing
 import os
 
 import torch
-import torch_npu  # noqa: F401
 from vllm.distributed.parallel_state import (get_world_group,
                                              init_distributed_environment)
 from vllm.utils import update_environment_variables