This allows the batch submitter to attempt to encode larger ranges of
transactions before they are compressed and tested against the max tx
size. Empirically this should give us ~3x throughput.