Type inference problem when tying a class with a type var

Please consider

import dataclasses as dtc
import typing as tp

T = tp.TypeVar('T', int, str, covariant=True)

@dtc.dataclass(frozen=True)
class C(tp.Generic[T]):
    c: type[T]

c_decision: dict[bool, C] = {
    True: C(int),
    False: C(str),
}

reveal_type(c_decision[True])
reveal_type(c_decision[False])

then

$ mypy -V
mypy 0.942
$ mypy issues-mypy/t9.py
issues-mypy/t9c.py:15: note: Revealed type is "t9c.C*[Any]"
issues-mypy/t9c.py:16: note: Revealed type is "t9c.C*[Any]"

If I remove type annotation for c_decision

c_decision = {
    True: C(int),
    False: C(str),
}

then the output is

issues-mypy/t9c.py:15: note: Revealed type is "builtins.object*"
issues-mypy/t9c.py:16: note: Revealed type is "builtins.object*"

EDIT: rewriting the description to try to simplify the problem

Start by running mypy on the shown issues-mypy/t9.py reproducer, or the referenced t9c.py variant, and compare revealed types with and without c_decision's annotation. Trace inference for C(int) and C(str); done when the intended behavior is captured by a regression test.