[FEA] Profiling tool auto-tuner should keep reducing the maxPartitionBytes if a table scan stage has heavy spilling and have OOM tasks

[viadea]

I wish Profiling tool auto-tuner should keep reducing the maxPartitionBytes if a table scan stage has heavy spilling and have OOM tasks.
This is proved to be working for at least one customer job.
My proposal is:

• It is a table scan stage
• It is spilling a lot(say some threashold)
• There are tasks in this stage failing with com.nvidia.spark.rapids.jni.CpuSplitAndRetryOOM: CPU OutOfMemory (Or whatever types of OOM such as GPU OOM, heap OOM, etc)
  Then each time auto-tuner keep reducing maxPartitionBytes into half.