Fix UncheckedTuples not failing for wrong amount of arguments

[Germandrummer92]

Currently apischema fails to correctly serialize a correctly typed dataclass with a "conflicting" union.

E.g.

from dataclasses import dataclass
from typing import Union, Tuple

import apischema


@dataclass(frozen=True)
class SomeTupleClass:
    bar: Union[Tuple[int, int], Tuple[int, int, int]]


def test_correct_serialization() -> None:
    serialized_dict = {
        "bar": [0, 0, 0]
    }

    as_python_object = apischema.deserialize(type=SomeTupleClass, data=serialized_dict)

    assert as_python_object == SomeTupleClass(bar=(0, 0, 0))

    assert apischema.serialize(as_python_object) == serialized_dict

currently fails as the serialization doesn't fail on trying to serialize into the first tuple.

It can be fixed by setting check_type=True, but I am not sure if this is really a "type" check or just the union serialization behaving wrongly.

For now we worked around it with check_type, but as your documentation says https://wyfo.github.io/apischema/0.18/de_serialization/#type-checking this has futher performance impacts, I would like to discuss if the tuple method shouldn't also be able to handle this conflicting union, as the typing in the dataclass itself seems correct

[Germandrummer92]

I guess python3.6 is not working anymore?

[wyfo]

I've just one night to release 0.18, so I'm a bit in a hurry, that's why I close your PR in favor of #521
(your name will appear in changelog, though)