In the AWS world, Athena does more or less what you are describing: but is is targeted (AFAIU) for large datasets. I'm not sure how it would perform on a single (small) CSV to shave off a few bytes in transit.