Fetchers docstring examples trigger dataset fetch in CI

[jeremiedbb]

Docstring examples were recently added to the fetchers. This makes the doc tests executed by pytest actually fetch the datasets.
In the fetcher tests we took some precaution to not fetch the real datasets, see

scikit-learn/sklearn/datasets/tests/test_lfw.py

Line 40 in 1bbb228

def mock_data_home(tmp_path_factory):

It has a significant impact on the duration of the test suite (and probably on memory usage as well)

27.75s call     ::__init__.py::_lfw.py::sklearn.datasets._lfw.fetch_lfw_pairs
20.98s call     ::__init__.py::_lfw.py::sklearn.datasets._lfw.fetch_lfw_people

[lesteve]

This is probably doable to skip doctests in sklearn/conftest.py pytest_collection_modifyitems based on SKLEARN_SKIP_NETWORK_TESTS.

[glemaitre]

Uhm normally, I almost sure to have written some code to avoid this using some fixture.

[glemaitre]

At list it was the intent behind this PR:

https://github.com/scikit-learn/scikit-learn/pull/28027/files#diff-3cdc9913d34276436e69cd506f11cdc77ef286069bd5465aaeeb73ac2bdee318

[glemaitre]

And indeed, the fetch_lfw_pairs is not added as a fixture here:

scikit-learn/sklearn/conftest.py

Lines 71 to 80 in 1bbb228

	dataset_fetchers = {
	"fetch_20newsgroups_fxt": fetch_20newsgroups,
	"fetch_20newsgroups_vectorized_fxt": fetch_20newsgroups_vectorized,
	"fetch_california_housing_fxt": fetch_california_housing,
	"fetch_covtype_fxt": fetch_covtype,
	"fetch_kddcup99_fxt": fetch_kddcup99,
	"fetch_olivetti_faces_fxt": fetch_olivetti_faces,
	"fetch_rcv1_fxt": fetch_rcv1,
	"fetch_species_distributions_fxt": fetch_species_distributions,
	}

So this is a simple fix.

/take