Function's argument type inference is affected by the functions return type when in a comprehension

Bug Report

MyPy seems to be getting confused if function that receives a generic class returns a union type with that generic is used in a comprehension. If the functions is used in a for loop or the function returns just the generic T, then MyPy doesn't show any errors.

To Reproduce

import dataclasses
from typing import Callable, TypeVar, Generic

T = TypeVar('T')


@dataclasses.dataclass(frozen=True, slots=True)
class Capsule(Generic[T]):
    builder: Callable[[], T]


def process(argument: Capsule[T]) -> T | RuntimeError:
    if isinstance((value := argument.builder()), int) and value % 2 != 0:
        return RuntimeError('Not an even number')

    return argument.builder()


encapsulated_ints = tuple(Capsule(builder=lambda: item) for item in range(10))

requests: list[int | RuntimeError] = []
for encapsulated_int in encapsulated_ints:
    request = process(encapsulated_int)
    requests.append(request)
requests = list(process(encapsulated_int) for encapsulated_int in encapsulated_ints)

Error:

main.py:25: error: Argument 1 to "process" has incompatible type "Capsule[int]"; expected "Capsule[int | RuntimeError]"  [arg-type]

https://mypy-play.net/?mypy=latest&python=3.12&gist=ffe7822cc564a6f347c2b8d6fd98c47f

Expected Behavior

The list comprehension line should be interpreted exactly the same as the for loop.

Actual Behavior

MyPy somehow relates the function return to the generic type.

Your Environment

• Mypy version used: 1.9.0
• Mypy command-line flags: no flags
• Mypy configuration options from mypy.ini (and other config files): no configs
• Python version used: 3.12

Start with the Python reproducer in the issue and compare the diagnostic for the for loop with the generator comprehension. Trace how generic argument inference uses the function return type in each case. Done means the comprehension is accepted like the loop and a regression test covers the reported example.