fix: swapped _R_co and _P to prevent type error. (#20508)

When loading a pytorch-lightning model from MLFlow, I get `TypeError: Type parameter +_R_co without a default follows type parameter with a default`.  This happens whenever doing `import pytorch_lightning as pl` which is done by packages like MLFlow.

Co-authored-by: Jirka Borovec <[email protected]>
Co-authored-by: Luca Antiga <[email protected]>
(cherry picked from commit ee7fa43)

Author: 3 people

src/lightning/pytorch/utilities/model_helpers.py

@@ -104,7 +104,7 @@ def _check_mixed_imports(instance: object) -> None:
 _R_co = TypeVar("_R_co", covariant=True)  # return type of the decorated method
 
 
-class _restricted_classmethod_impl(Generic[_T, _P, _R_co]):
+class _restricted_classmethod_impl(Generic[_T, _R_co, _P]):
     """Drop-in replacement for @classmethod, but raises an exception when the decorated method is called on an instance
     instead of a class type."""