A practical FinOps playbook for Databricks. Cluster types, file compaction, caching, serverless, and BI rationalization — with realistic savings ranges.