FEAT Add make_series and make_dataframe

[Vincent-Maladiere]

Addresses https://github.com/skrub-data/skrub/pull/784/files#r1358253388

This simple PR implements, for both Pandas and Polars:

• A function to create a dataframe from a dictionary of columns
• A function to create a series from a 1d array

When merged, this can be used directly in #784 for example.

[jeromedockes]

what kind of mistake do you foresee to make this check necessary?

[Vincent-Maladiere]

It may be slightly overkill, but I think about dictionaries of 2d arrays or dataframes.
For instance:

pd.DataFrame(
	dict(
		a=[[1, 2, 1]],
		b=[1, 2, 3],
	)
)

will raise ValueError: All arrays must be of the same length, but this error is not informative enough IMO.
I suspect this kind of error might happen more than we'd expect. WDYT?

[jeromedockes]

but this is a private function, right? that will be used only by skrub code not directly by users. so if skrub passes it incorrect inputs there isn't too much the user can do about it except report the problem

btw WDYT about renaming dataframe to _dataframe? we want skrub estimators to support both pandas and polars, but I'm not sure we want to publicly expose the machinery that makes it possible

[Vincent-Maladiere]

Ok for the renaming, I'll remove the check

[Vincent-Maladiere]

I adjusted the API to make the dataframe module private (it seems I still have some work on the examples).
I also did some housekeeping by removing the DataFrameLike and SeriesLike type hints, LMKWYT @jeromedockes :)

[jeromedockes]

LGTM, I think once you remove the merge conflict we can merge it

[jeromedockes]

I think you also need to update test_select_cols

[jeromedockes]

and the docstring of get_df_namespace also mentions skrub.dataframe

[jeromedockes]

LGTM, thanks!