xgboost: resolve scikit-learn from the registry instead of compiling it

scikit-learn was deliberately absent from PIP_ONLY_BINARY because no riscv64
wheel existed, so pip fell back to its sdist and spent about 22 minutes of a
63 minute job compiling it in-container. #339 is merged and publishing 1.9.0,
which is the version pip resolves to here, so adding it to the list forces the
registry wheel and removes that compile.

The test phase will fail until the publish lands: with scikit-learn in
PIP_ONLY_BINARY there is no sdist fallback left, and xgboost.testing
importorskips sklearn.datasets at module level, so the suite needs it.

Author: luhenry

.github/workflows/build-xgboost.yml

@@ -72,12 +72,12 @@ jobs:
         env:
           CIBW_MANYLINUX_RISCV64_IMAGE: ${{ env.MANYLINUX_RISCV64_IMAGE }}
           CIBW_ENVIRONMENT: PIP_EXTRA_INDEX_URL=https://pypi.riseproject.dev/simple/
-          # numpy/scipy/pandas resolve to our registry's riscv64 wheels rather
-          # than to newer sdists on public PyPI. scikit-learn is deliberately
-          # left out of the list: it has no riscv64 wheel anywhere yet, and the
-          # whole suite skips without it (xgboost.testing importorskips
-          # sklearn.datasets at module level), so it is compiled in-container.
-          CIBW_TEST_ENVIRONMENT: PIP_ONLY_BINARY=numpy,scipy,pandas
+          # Every scientific dependency resolves to our registry's riscv64 wheels
+          # rather than to a newer sdist on public PyPI. scikit-learn is in the
+          # list now that we publish it; without it pip compiled it in-container,
+          # which the whole suite needs because xgboost.testing importorskips
+          # sklearn.datasets at module level.
+          CIBW_TEST_ENVIRONMENT: PIP_ONLY_BINARY=numpy,scipy,pandas,scikit-learn
           CIBW_TEST_REQUIRES: pytest hypothesis joblib pandas scikit-learn
           CIBW_TEST_SOURCES: tests/pytest.ini tests/python demo/data
           CIBW_TEST_COMMAND: >-