fix: LRL1 results now reproducible

[immu4989]

Why are these changes needed?

Summary

• Seed random_state on LRL1Classifier so LogisticRegression(solver="saga", penalty="l1") produces deterministic results across runs.
• Uses the same defensive pattern as fix: SGD results now reproducible #1541: pop the FLAML-internal random_seed key from self.params, and only set random_state when the caller has not already provided one.
• Uncomment "lrl1" in both classification reproducibility test parametrize lists (it was previously disabled).

Why

LRL1Classifier defaults to solver="saga", a stochastic-gradient solver that shuffles samples each pass. Without random_state, identical fits produce different results — same root cause as SGD (#1541). LRL2Classifier is unaffected since it uses the deterministic lbfgs solver.

Test plan

• pytest test/automl/test_classification.py -k "reproducibility and lrl1" — both wrapper and underlying-model tests pass
• pre-commit run --files flaml/automl/model.py test/automl/test_classification.py — all hooks pass
• CI green on the PR

Related issue number

Follows the same pattern as LGBM (#1369), CatBoost (#1364), ElasticNet (#1374), LinearSVC (#1376), and SGD (#1541).

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.

[Copilot]

Pull request overview

This PR improves determinism for FLAML’s lrl1 estimator by ensuring LRL1Classifier seeds scikit-learn’s stochastic LogisticRegression(solver="saga", penalty="l1"), and re-enables reproducibility coverage for lrl1 in the classification test suite.

Changes:

• Set LRL1Classifier.params["random_state"] from FLAML’s internal random_seed (defaulting to 10242048) when the caller hasn’t explicitly provided random_state.
• Remove (pop) random_seed from LRL1Classifier.params to avoid passing an unsupported parameter into scikit-learn constructors.
• Re-enable lrl1 in both classification reproducibility parametrized test lists.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
flaml/automl/model.py	Seeds LRL1Classifier’s underlying LogisticRegression via random_state for deterministic saga behavior.
test/automl/test_classification.py	Re-enables lrl1 in reproducibility test parametrizations to prevent regressions.