fix: use dtype instead of deprecated torch_dtype

[xyf5432]

Summary

config.torch_dtype and the torch_dtype keyword argument were deprecated in transformers 4.56 (PR #39782) and replaced by dtype. This PR updates the AutoModelForSequenceClassification.from_pretrained call in contrib/rlhflow/reward_modeling.py (line 44) to select the keyword from the installed transformers version using packaging.version, so transformers < 4.56 keeps working.

Changes

• contrib/rlhflow/reward_modeling.py: add a _dtype_kwargs helper that returns {"dtype": ...} on transformers >= 4.56 and {"torch_dtype": ...} otherwise; the from_pretrained call passes it as **_dtype_kwargs(torch.bfloat16).

Test

• The modified file compiles (python -m py_compile).
• On transformers >= 4.56 the call uses dtype and emits no deprecation warning; on older versions it passes torch_dtype unchanged.

Fixes #977