(🐞) No error for incorrect variance on parameter with generic type

from typing import TypeVar, Generic

T_in = TypeVar("T_in", contravariant=True)
T_out = TypeVar("T_out", covariant=True)

class AOut(Generic[T_out]): ...
class AIn(Generic[T_in]): ...

class BOut(Generic[T_out]):
    def f(self, a_out: AOut[T_out]) -> None: ...  # no error
    
class BIn(Generic[T_in]):
    def f(self, a_out: AIn[T_in]) -> None: ...  # no error

Mypy in real life

from typing import TypeVar, Generic, Callable

T_in = TypeVar("T_in", contravariant=True)
T_out = TypeVar("T_out", covariant=True)

class AOut(Generic[T_out]):
    T: T_out
    def f(self, fn: Callable[[], T_out]) -> None:
        self.t = fn()
    
    def dump(self) -> T_out:
        return self.t
class AIn(Generic[T_in]):
    t: T_in
    def load(self, t: T_in) -> None:
        self.t = t
    def f(self, fn: Callable[[T_in], object]) -> None:
        fn(self.t)
    

o1: AOut[int]
o: AOut[object] = o1

o.f(lambda: "")
o1.dump() + 1  # runtime error

i1: AIn[object]
i: AIn[int] = i1

i1.load("")
i.f(lambda x: x + 1)  # runtime error

related: #734, #8191

Run mypy on the two reproductions in the issue and confirm that the invalid variance uses produce no errors. Trace the generic variance-checking entry point and add a regression test covering both covariant and contravariant cases; done means mypy reports the expected errors without breaking valid variance usage.