Fails to infer Literal when required by return type

Bug Report

When a return type requires a Literal, the value in a return type does not seem to infer a Literal when the code is correct.

To Reproduce

This could be reduced further, but this should fully demonstrate what I'm trying to achieve, in case there are any other intricacies:

from typing import Generic, Literal, TypeVar, overload

_T = TypeVar("_T")
_Code = TypeVar("_Code", bound=int, default=Literal[200])

class APIResponse(Generic[_T, _Code]):
    @overload
    def __init__(self, body: _T):
        ...
    @overload
    def __init__(self, body: _T, status: _Code):
        ...
    def __init__(self, body: _T, status: int = 200):
        ...


def foo(b: bool) -> APIResponse[int, Literal[200]] | APIResponse[str, Literal[404]]:
    if b:
        return APIResponse(5)
    return APIResponse("FOO", status=404)  # return-value error

Expected Behavior

While it may be difficult to infer Literal everywhere, I expected mypy to correctly infer the type when defined in the return statement. Atleast, mypy seems to do this for TypedDict.

Start by reproducing the supplied APIResponse example in mypy and compare its behavior with the TypedDict case mentioned in the report. Trace the return-value checking and generic type inference involved in the return statement; done means the shown function type-checks without the return-value error while preserving the declared Literal alternatives.