WIP - improve dataset fetchers

[rcap107]

This PR is improving the dataset fetcher functions. It addresses #1422 by adding the path to the dataset file to the Bunch object returned by the fetcher.

I am also adding the default data folder of skrub to the configuration file, and I'm deprecating the original name SKRUB_DATA_DIRECTORY in favor of SKB_DATA_DIRECTORY, to follow the same format as the other environment variables set by skrub.

[rcap107]

Something I did not consider is that some datasets have multiple paths (such as fetch_plane_delays).

For the moment I am returning a list of paths, but then the result is that then single datasets become clunky to load, like

data = fetch_employee_salaries()
df = pd.read_csv(data["paths"][0])

I'm not sure what's the best way to deal with this