[2.0 regression] TypeVar with constrains Sub and Super does not accept Sub as a default value

[hauntsaninja]

λ cat test.py   
from typing import TypeVar
T = TypeVar("T", int, float)
def f(x: T = 0): ...

λ mypy test.py --no-incremental                       
test.py:3: error: Incompatible default for parameter "x" (default has type "int", parameter has type "T")  [assignment]
Found 1 error in 1 file (checked 1 source file)

Bisects to #21119

[ilevkivskyi]

This is not specific to int vs float, i.e. not related to promotions. The same reproduces with:

class B: ...
class C(B): ...

S = TypeVar("S", B, C)
c = C()
def g(x: S = c): ...