Fix inferred type of is_dataclass(Any)

[intgr]

Fixes #12401 by adding a hack, in the form of overload def is_dataclass(obj: Never) -- because Any is compatible with Never, but no real type is.

Previously is_dataclass(Any) matched the wrong overload rule and inferred incorrect return type.

[intgr]

Suppressing mypy error:

stdlib/dataclasses.pyi:219: error: Narrowed type "DataclassInstance | type[DataclassInstance]" is not a subtype of input type "Never" [narrowed-type-not-subtype]

[intgr]

Actually these cases are already checked by check_other_isdataclass_overloads below and can be removed.

Suggested change

	def is_dataclass_object(arg: object) -> None:
	if dc.is_dataclass(arg):
	assert_type(arg, Union["DataclassInstance", Type["DataclassInstance"]])
	def is_dataclass_type(arg: type) -> None:
	if dc.is_dataclass(arg):
	assert_type(arg, Type["DataclassInstance"])

[srittau]

The primer output looks promising. It mostly changes wrong error messages into true positives. (Due to is_dataclass()'s surprising API.)

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

pydantic (https://github.com/pydantic/pydantic)
+ pydantic/v1/json.py:80: error: Argument 1 to "asdict" has incompatible type "DataclassInstance | type[DataclassInstance]"; expected "DataclassInstance"  [arg-type]
- pydantic/v1/json.py:80: error: No overload variant of "asdict" matches argument type "type[DataclassInstance]"  [call-overload]
- pydantic/v1/json.py:80: note: Possible overload variants:
- pydantic/v1/json.py:80: note:     def asdict(obj: DataclassInstance) -> dict[str, Any]
- pydantic/v1/json.py:80: note:     def [_T] asdict(obj: DataclassInstance, *, dict_factory: Callable[[list[tuple[str, Any]]], _T]) -> _T

pytest (https://github.com/pytest-dev/pytest)
+ src/_pytest/_io/pprint.py:114: error: Unused "type: ignore" comment  [unused-ignore]
+ src/_pytest/_io/pprint.py:116: error: "DataclassInstance" has no attribute "__dataclass_params__"  [attr-defined]
+ src/_pytest/_io/pprint.py:122: error: Unused "type: ignore" comment  [unused-ignore]

core (https://github.com/home-assistant/core)
+ homeassistant/util/frozen_dataclass_compat.py:114: error: Argument 1 to "__new__" of "object" has incompatible type "DataclassInstance | type[DataclassInstance]"; expected "type[DataclassInstance]"  [arg-type]

streamlit (https://github.com/streamlit/streamlit)
+ lib/streamlit/runtime/caching/hashing.py:409:53: error: Argument 1 to "asdict" has incompatible type "Union[DataclassInstance, Type[DataclassInstance]]"; expected "DataclassInstance"  [arg-type]
- lib/streamlit/runtime/caching/hashing.py:409:34: error: No overload variant of "asdict" matches argument type "Type[DataclassInstance]"  [call-overload]
- lib/streamlit/runtime/caching/hashing.py:409:34: note: Possible overload variants:
- lib/streamlit/runtime/caching/hashing.py:409:34: note:     def asdict(obj: DataclassInstance) -> Dict[str, Any]
- lib/streamlit/runtime/caching/hashing.py:409:34: note:     def [_T] asdict(obj: DataclassInstance, *, dict_factory: Callable[[List[Tuple[str, Any]]], _T]) -> _T

koda-validate (https://github.com/keithasaurus/koda-validate)
+ koda_validate/typehints.py:119: error: Argument 1 to "DataclassValidator" has incompatible type "DataclassInstance | type[DataclassInstance]"; expected "type[Any]"  [arg-type]
+ koda_validate/signature.py:81: error: Argument 1 to "DataclassValidator" has incompatible type "DataclassInstance | type[DataclassInstance]"; expected "type[Any]"  [arg-type]
+ koda_validate/signature.py:81: error: Argument 1 to "dataclass_no_coerce" has incompatible type "DataclassInstance | type[DataclassInstance]"; expected "type[DataclassInstance]"  [arg-type]

xarray-dataclasses (https://github.com/astropenguin/xarray-dataclasses)
- xarray_dataclasses/typing.py:300: error: Unused "type: ignore" comment  [unused-ignore]

[srittau]

LGTM, but I'd like a third opinion.

[max-muoto]

Thoughts on adding a TODO to revert things here? Once python/mypy#17579 is fixed. At least making it clear in the stubs this is a MyPy specific issue we're accounting for.