ENH polars support for InterpolationJoiner

[TheooJ]

Closes #897

>>> import polars as pl
>>> from sklearn.neighbors import KNeighborsRegressor
>>> from skrub import InterpolationJoiner

>>> buildings = pl.DataFrame(
...     {"latitude": [1.0, 2.0], "longitude": [1.0, 2.0], "n_stories": [3, 7]}
... )
>>> annual_avg_temp = pl.DataFrame(
...     {
...         "latitude": [1.2, 0.9, 1.4, 1.7, 5.0],
...         "longitude": [0.8, 1.1, 1.8, 1.8, 5.0],
...         "avg_temp": [10.0, 11.0, 15.0, 16.0, 20.0],
...     }
... )

>>> joiner = InterpolationJoiner(
...     annual_avg_temp,
...     key=["latitude", "longitude"],
...     regressor=KNeighborsRegressor(2),
... )
>>> joiner.fit_transform(buildings)

┌──────────┬───────────┬───────────┬──────────┐
│ latitude ┆ longitude ┆ n_stories ┆ avg_temp │
│ ---      ┆ ---       ┆ ---       ┆ ---      │
│ f64      ┆ f64       ┆ i64       ┆ f64      │
╞══════════╪═══════════╪═══════════╪══════════╡
│ 1.0      ┆ 1.0       ┆ 3         ┆ 10.5     │
│ 2.0      ┆ 2.0       ┆ 7         ┆ 15.5     │
└──────────┴───────────┴───────────┴──────────┘

As discussed with @jeromedockes, if the joined dataframe could contain duplicate column names, we add random suffixes and don't raise -- to be consistent with the behavior of the other joiners

[jeromedockes]

a first pass

[jeromedockes]

here why is this check not necessary anymore?

[TheooJ]

We were just fitting on two different column names ("A" and "rB"). Now the test checks that the double fitting returns unique column names. Other tests check fitting on left + right keys

[jeromedockes]

why do we remove this test?

[TheooJ]

The case where we join [0, 1] on [1, 0] is tested in test_fit_on_none. I can separate into two tests if necessary

[TheooJ]

I readded a test (for pandas only). Note that because of concat_horizontal, the initial index is reset. We could store it and reapply it in the case of pandas

It's possible to store the index of the first arg of concat_horizontal and set the output to have this index, WDYT ?

[jeromedockes]

@TheooJ could you do git commit --allow-empty -m '[doc build]' and push? so that we can see the rendered example

[jeromedockes]

thanks a lot @TheooJ I think it's almost ready

[jeromedockes]

very cool, thanks @TheooJ !