shapely: Fix tuple length of CoordinateSequence items

[hamdanal]

Closes #13422

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[Avasam]

How about making this a TypeVar? I can't find a way to encode this directly in the __init__ using the np.ndarray type (ie: extracting the shape size into a tuple length), but you could do this using TypeVar default:

_CoordinateTuple_co = TypeVar("_CoordinateTuple_co", covariant=True, bound=tuple[float, ...], default=tuple[float, ...])

class CoordinateSequence(Generic[_CoordinateTuple_co]):
    def __init__(self, coords: NDArray[np.float64]) -> None: ...
    def __len__(self) -> int: ...
    def __iter__(self) -> Iterator[_CoordinateTuple_co]: ...
    @overload
    def __getitem__(self, key: int) -> _CoordinateTuple_co: ...
    @overload
    def __getitem__(self, key: slice) -> list[_CoordinateTuple_co]: ...
    def __array__(self, dtype: DTypeLike | None = None, copy: Literal[True] | None = None) -> NDArray[np.float64]: ...
    @property
    def xy(self) -> tuple[array[float], array[float]]: ...

Which makes it possible to "opt-in" stricter/more accurate length-checks.

my_array: NDArray[np.float64] = ...
# revealed type: CoordinateSequence[tuple[float, ...]]
test = CoordinateSequence(my_array)
# revealed type: CoordinateSequence[tuple[float, float]]
test2 = CoordinateSequence[tuple[float, float]](my_array)
# revealed type: CoordinateSequence[tuple[float, float, float]]
test3 = CoordinateSequence[tuple[float, float, float]](my_array)

Otherwise looks good to me.

[hamdanal]

Thanks for the review Avasam.

CoordinateSequence is an internal class that should not be initialized directly by the user. Also, we don't currently keep track of whether shapely objects have a z coordinate or not anywhere in shapely, so it would be useless to have this class generic if we can never specialize it (also I use shapely extensively and never had a use case for it).
Using tuple[float, ...] works very well because both mypy and pyright will happily allow you to unpack it to x, y or x, y, z. Do you have a use case for what you proposed?

[Avasam]

Thank you for the additional details! I don't use shapely so my comment was entirely based on looking at the implementation. With your additional comments in mind, I think your PR is completely fine as-is.

If someone actually needs to make the distinction on the tuple size, then they can request it.

This is a very simple PR so I'll merge now. I don't expect pushback from other maintainers on this.