Native Iceberg writes for CTAS / RTAS on Spark 3.4 #7428

What is the problem the feature request solves?

On Spark 3.4, CREATE TABLE AS SELECT and REPLACE TABLE AS SELECT into Iceberg plan AtomicCreateTableAsSelectExec / AtomicReplaceTableAsSelectExec, and the staged table writes inline. These writes are not eligible for native Iceberg writes, so the write and the operators above the last shuffle stay in Spark.

Describe the potential solution

Plan the staged CTAS/RTAS write on Spark 3.4 through the same native Iceberg write path used for AppendData, and commit through the staged table.

Additional context

The Iceberg guide notes that CTAS/RTAS on Spark 3.4 writes inline today.