feat: Implement basic `train_test_split`

[augustebaum]

This PR implements the train_test_split function which wraps around sklearn.

The warnings are only emitted using warnings.warn for now; nothing is saved to a Project, as that part of the design still needs work.

In order to validate the technical design one warning is implemented, HighClassImbalanceWarning. This inherits from the built-in Warning class so that it can be used directly in warnings.warn.

Because we expect warnings to take a wide range of arguments, we pack all the relevant information into a kwargs dict inside train_test_split, and let the check static method accept **kwargs as well as the necessary positional arguments.

Closes https://github.com/probabl-ai/skore/issues/683

---

Todo:

• [ ] Add an example
• [ ] Test behaviour in a Jupyter notebook
  • [ ] Make warnings display nicely in Jupyter notebook: currently they show as unwrapped lines of text (so you need to scroll a lot to read the messages)
• [x] Deal with the case of train_test_split(X, test_size=...); previously we took arrays[-1] as y, but now we only do this if there are at least 2 arrays
• [x] Refactor _find_ml_task, as a function of the same name already exists in cross_validate.py?

[sylvaincom]

Thanks, I got some issues:

from skore.sklearn.train_test_split import train_test_split

• [x] should be from skore import train_test_split

X = [[1]] * 4
y = [0, 1, 1, 1]
train_test_split(X, y)

• [ ] the warning message does not return to line in VS Code:

https://github.com/user-attachments/assets/80845d61-8dfc-40d4-8307-84b4701e4c80