When designing Databricks workflows, which pattern is best suited for managing data as it progresses through various stages of refinement, from its raw form to being ready for use?