Fix typing error complains due to upstream torch.optimizer decorator …

…typing changes

Summary:
Due to pytorch/pytorch#144161, the current disabled `state_dict()` and `load_state_dict()` typing starts complains due to the incompatible between `typing.NoReturn` and `None` and `StateDict`.

Although `NoReturn` is a better typing for the disabled checkpoint solution, unfortunately this is not feasible due to `torch.optimizer` typing requirement.

Differential Revision: D67951563

distributed_shampoo/distributed_shampoo.py

@@ -12,7 +12,7 @@
 from collections.abc import Callable, Iterator, Sequence
 from copy import deepcopy
 from functools import partial
-from typing import Any, NoReturn
+from typing import Any
 
 import torch
 
@@ -1098,12 +1098,12 @@ def step(self, closure: Callable[[], float] | None = None) -> float | None:  # t
 
         return loss
 
-    def state_dict(self) -> NoReturn:
+    def state_dict(self) -> StateDict:
         raise NotImplementedError(
             "Distributed Shampoo does not support the standard state_dict() method for checkpointing!"
         )
 
-    def load_state_dict(self, state_dict: StateDict) -> NoReturn:
+    def load_state_dict(self, state_dict: StateDict) -> None:
         raise NotImplementedError(
             "Distributed Shampoo does not support the standard load_state_dict() method for checkpointing!"
         )