Dispatch for xpu in adaptive_avg_pooling (#84541)
Motivation:
- See native_functions.yaml, operators adaptive_avg_pool2d/adaptive_avg_pool3d are not recommended to register for backends.
- When adaptive_avg_pool2d/adaptive_avg_pool3d have a input of xpu tensor, they can't step into xpu implementation.
Solution:
- Dispatch to _adaptive_avg_pool2d/_adaptive_avg_pool3d for xpu backend in adaptive_avg_pool2d/adaptive_avg_pool3d implementation.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84541
Approved by: https://github.com/ezyang
