Fix MLflow autologged model reload fitted-state detection

[Copilot]

Recent scikit-learn versions treat MLflow-reloaded FLAML pipelines as unfitted because FLAML estimator wrappers do not expose their fitted state through sklearn’s protocol. Consequently, loaded.predict() raises NotFittedError.

• Estimator compatibility

  • Implement __sklearn_is_fitted__() based on FLAML’s underlying _model state.
  • Preserve the existing MLflow sklearn pipeline artifact format.

• Regression coverage

  • Log and reload an autologged FLAML model through runs:/{run_id}/model.
  • Verify sklearn recognizes the pipeline as fitted and its predictions match AutoML.

loaded = mlflow.sklearn.load_model(f"runs:/{run_id}/model")
loaded.predict(X_test)