[FSDP][state_dict] Make sharded_state_dict work with composable fully_shard (#100856)
The current implementation of sharded_state_dict only works with wrapper based FSDP (both use_orig_params and not use_orig_params work) but not with fully_shard. This PR changes the implementation of sharded_state_dict when loading to fix the incompatibility.
Differential Revision: [D45626856](https://our.internmc.facebook.com/intern/diff/D45626856/)
**NOTE FOR REVIEWERS**: This PR has internal Meta-specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D45626856/)!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100856
Approved by: https://github.com/awgu, https://github.com/zhaojuanmao