[FSDP] Full precision in eval mode (#97645)
If model.eval() is true, then runs the model in full precision.
Changes:
- Changed _force_full_precision to check self.is_training
- Check for _force_full_precision when casting gradients to reduced dtype
- Small change when accessing _full_prec_param_padded
- tests for class based and fully_shard APIs
Differential Revision: [D43933690](https://our.internmc.facebook.com/intern/diff/D43933690/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97645
Approved by: https://github.com/awgu
