fix: no pinning memory on CPU (deepmodeling#4874)

This PR fixes the warning message when running on CPU:
```
torch/utils/data/dataloader.py:665: UserWarning: 'pin_memory' argument is set as true but no accelerator is found, then device pinned memory won't be used.
  warnings.warn(warn_msg)
  ```

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

## Summary by CodeRabbit

* **Bug Fixes**
  * Improved memory handling by enabling memory pinning only when using non-CPU devices, which may enhance performance and stability when training on GPUs.

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

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: 2 people

deepmd/pt/train/training.py

@@ -208,7 +208,7 @@ def get_dataloader_and_iter(_data, _params):
                     else 0,  # setting to 0 diverges the behavior of its iterator; should be >=1
                     drop_last=False,
                     collate_fn=lambda batch: batch,  # prevent extra conversion
-                    pin_memory=True,
+                    pin_memory=(DEVICE != "cpu"),  # pin memory only if not on CPU
                 )
                 _data_iter = cycle_iterator(_dataloader)
                 return _dataloader, _data_iter