[tests][python-package] Modify tests using boston housing dataset

[jmoralez]

Description

It was recently brought to light that the boston housing dataset has an ethical problem (which is more thoroughly described in scikit-learn/scikit-learn#16155) and thus will be removed from scikit-learn in version 1.2 (Olivier Grisel's explanation). Since some of the tests for the python package rely on this dataset (sample logs with the warning) they should be changed to use a different dataset.

Tests currently using the boston dataset:

• test_engine::test_regression
• test_engine::test_continue_train
• test_engine::test_continue_train_reused_dataset
• test_engine::test_continue_train_dart
• test_engine::test_cv
• test_engine::test_feature_name
• test_engine::test_save_load_copy_pickle
• test_engine::test_mape_rf
• test_engine::test_mape_dart
• test_engine::test_model_size
• test_engine::test_get_split_value_histogram
• test_engine::test_early_stopping_for_only_first_metric
• test_engine::test_extra_trees
• test_engine::test_path_smoothing
• test_engine::test_interaction_constraints
• test_engine::test_predict_with_start_iteration
• test_engine::test_force_split_with_feature_fraction
• test_sklearn::test_regression
• test_sklearn::test_objective_aliases
• test_sklearn::test_regression_with_custom_objective
• test_sklearn::test_dart
• test_sklearn::test_stacking_regressor
• test_sklearn::test_clone_and_property
• test_sklearn::test_joblib
• test_sklearn::test_non_serializable_objects_in_callbacks
• test_sklearn::test_evaluate_train_set
• test_sklearn::test_metrics
• test_sklearn::test_first_metric_only
• test_sklearn::test_training_succeeds_when_data_is_dataframe_and_label_is_column_array

[StrikerRUS]

Thanks for writing this up!
I just converted bulleted list into TODO list to track the progress.

Also, we should stop accept PRs with tests that utilizes this dataset.