What are the best optimization techniques for loading large CSV data to ORC format in Spark SQL