A data engineer runs a nightly Spark notebook in Fabric that joins several large Delta tables. In the Spark UI, they notice many stages create hundreds of tiny output partitions after shuffles, and some joins that could use a smaller build side are executing as sort-merge joins. The engineer wants Spark to dynamically coalesce shuffle partitions and switch join strategies at runtime based on actual data statistics, without manually tuning partition counts. Which Spark capability should they ensure is enabled?