[FSDP][optim_state_dict] Move local optimizer state to FSDP compute_device (#110929)
This will ensure all the tensors are on FSDP compute_device.
Differential Revision: [D50059492](https://our.internmc.facebook.com/intern/diff/D50059492/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/110929
Approved by: https://github.com/wz337