Currently projections is not passed down to storage engines, therefore we are inputing all data from the data source where it is unnecessary.
Passing down projections can allow the storage engine to only select the fields necessary and potentially reduce each batch size