fix: LRL1/LRL2 — migrate off sklearn 1.10 deprecations (penalty, n_jobs)

[immu4989]

Why are these changes needed?

LRL1Classifier and LRL2Classifier configure sklearn's LogisticRegression with two arguments that were deprecated in scikit-learn 1.8 and scheduled for removal in scikit-learn 1.10:

• penalty= (use l1_ratio or C instead — l1_ratio=1 for L1, l1_ratio=0 for L2)
• n_jobs= (has no effect since 1.8 — SGDEstimator.config2params already pops this for the regression case)

Each lrl1 / lrl2 fit on sklearn ≥ 1.8 currently emits a FutureWarning for penalty, a UserWarning for penalty=l1 with l1_ratio=0.0, and a FutureWarning for n_jobs. Once scikit-learn 1.10 ships, these become TypeErrors and the two estimators hard-break.

This PR version-gates the configuration in config2params using the existing SKLEARN_VERSION constant in flaml/automl/model.py (already in use for the SGD loss migration at lines 2383 and 2437):

• LRL1Classifier: on sklearn ≥ 1.8, set l1_ratio=1.0 and drop n_jobs; on older sklearn, keep penalty="l1".
• LRL2Classifier: on sklearn ≥ 1.8, just drop n_jobs (the default LogisticRegression penalty is already "l2"); on older sklearn, keep penalty="l2".

Verified locally on sklearn 1.8.0: both migrations produce identical coef_ to the legacy configuration, the existing lrl1 / lrl2 reproducibility tests continue to pass, and no FutureWarning / UserWarning about penalty or n_jobs is emitted during fit.

Related issue number

Closes #1555

Checks

• I've used pre-commit to lint the changes in this PR (note the same in integrated in our CI checks).
• I've included any doc changes needed for https://microsoft.github.io/FLAML/. See https://microsoft.github.io/FLAML/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[thinkall]

The ci errors are caused by openml dataset downloading. Re-running.