Support out_dtype in torch.bmm

Since torch 2.8, `torch.bmm` supports an `out_dtype` parameter. From the [docs](https://docs.pytorch.org/docs/2.8/generated/torch.bmm.html) "the dtype of the output tensor, Supported only on CUDA and for torch.float32 given torch.float16/torch.bfloat16 input dtypes"

Could this be added?
