Move some broadcasting logic away from codegen. (#32982)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32982
For masked_scatter_ and masked_fill_ (which already have manually written wrappers), move the broadcasting logic into the manually written wrappers.
Test Plan: Imported from OSS
Differential Revision: D19726830
Pulled By: gchanan
fbshipit-source-id: 1f6e55e19c1314a76e43946b14d58f147c0f8204