Support for arbitraty generics

[rambip]

This is my attempt to fix #415

To be clear:

• I was in a rush, the solution was LLM-generated
• I did not review the code
• I only tested on my toy-example

If you consider it's slop and close it, I will have no problem with that.

Have a nice day !

[rambip]

Hum, it does not seem to solve this case right now:
now it does

def accuracy(model: nn.Module, loader: DataLoader[tuple[Tensor, Tensor]]) -> float:
    """Compute accuracy on a dataset.

    pre: loader is not None and len(loader) > 0
    inv: 0 <= return_value <= 1
    """
    correct = 0
    total = 0
    for X, y in loader:
        preds = model(X).argmax(dim=1)
        correct += (preds == y).sum().item()
        total += len(y)
    return correct / total

[pschanely]

Not slop, as far as I can tell. Thank you for the contribution!
I have a few (hopefully small) requests; feel free to add yourself at the bottom of contributing.rst too.

[rambip]

Is that ok now ?

[pschanely]

Is that ok now ?

Just a few more mypy typing issues, but otherwise this looks great!
FWIW, in many cases, I just remove the -> None from the test definition to disable checking for tests that are painful to get compliant.

[rambip]

I'm confused about the mypy errors, some seems unrelated to my change.

[pschanely]

I'm confused about the mypy errors, some seems unrelated to my change.

Unfortunately there are a lot of warning messages in there (I'll file an issue about this presently); for now, you only need to look at the errors.

[rambip]

Only hypothesis failing

[pschanely]

LGTM! Thank you!