[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>

examples/dp_offline/data_parallel.py

@@ -30,7 +30,6 @@ def main():
         for i in range(local_rank * tp_size, (local_rank + 1) * tp_size))
 
     import torch
-    import torch_npu  # noqa
     from vllm import LLM, SamplingParams
     from vllm.distributed.parallel_state import (
         destroy_distributed_environment, destroy_model_parallel)