Share your results benchmarking PySpark vs Scala Spark memory consumption on cloud clusters. What setups, configs, and OOM patterns have you seen in production?