In the context of a machine learning project with large-scale datasets, which Databricks MLlib supported technique is most effective for sampling and processing data efficiently for model training?