Added missing set_device calls for nvidia_transform

[Serge45]

In HIP_ENVIRONMENT is True case, there's missing calls of pre_call and post_call for proper device setting for nvidia_transform.

[lcskrishna]

@pnunna93 @Serge45 Please verify if we need to add the corner cases when state=none and out=none similar to without HIP. If there is no need, the PR is fine for merge.

[pnunna93]

They are present in nvidia_transform - https://github.com/Serge45/bitsandbytes/blob/feature/miss-set-device-transform/bitsandbytes/functional.py#L516-L520