A data engineering team needs to address failures within a multi-task Databricks Jobs workflow. To optimize resource utilization and time, they must repair failed tasks while ensuring minimal recomputation of tasks that have already completed successfully. What is the most efficient approach?