[inductor] fix cpp legalize bf16 reduction (#97228)
When legalizing bf16 for reduction, operators with result dtype of torch.int64, like argmax, would encounter an assertion error now. The PR fixes for the case of int64, enabling several bf16 models (hf_Reformer, doctr_reco_predictor) to run successfully.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97228
Approved by: https://github.com/jgong5, https://github.com/EikanWang, https://github.com/desertfire
