3 ways to run high-performance Apache Spark jobs on Apache Iceberg data