Fix signature codec with tuple type annotations

[sauerburger]

Fixes #1036

I'm unsure how/where to add a test for this as I didn't manage to run the test suite. Locally, I've experimented with the following snippet.

from typing import Tuple
import numpy as np
from mlserver.codecs.decorator import SignatureCodec

def predict_fn_typing_tuple() -> Tuple[np.ndarray, np.ndarray]:
    return (np.array([4]), np.array([2]))


def predict_fn_native_tuple() -> tuple[np.ndarray, np.ndarray]:
    return (np.array([4]), np.array([2]))


def predict_fn_tuple_object() -> (np.ndarray, np.ndarray):
    return (np.array([4]), np.array([2]))


def test_typing_tuple():
    signature_codec = SignatureCodec(predict_fn_typing_tuple)
    assert signature_codec._output_hints == [np.ndarray, np.ndarray]


def test_native_tuple():
    signature_codec = SignatureCodec(predict_fn_native_tuple)
    assert signature_codec._output_hints == [np.ndarray, np.ndarray]


def test_tuple_object():
    signature_codec = SignatureCodec(predict_fn_tuple_object)
    assert signature_codec._output_hints == [np.ndarray, np.ndarray]

Please note: The notation for the predict_fn_native_tuple test requires Python 3.9. The PR code, however, does not depend on Python 3.9.

[adriangonz]

Nice one! Thanks a lot for following up with this PR @sauerburger .

Changes look great! 🚀

For the tests, perhaps you can add a new one for the _as_list method, covering the different ways to specify the type hint? We run tests in 3.8 - 3.10, so we'll need to skip the tuple[np.ndarray, np.ndarray] form unfortunately. Happy to help out writing the test if you've got any questions.

[adriangonz]

Nice one @sauerburger ! Thanks for making those changes.

PR looks great. Once tests are green, it should be good to go ahead. 👍

[sauerburger]

Thanks for the kind feedback. I've fixed the issue that triggered the linter.