This is just a temporary remedy to (hopefully) make the devnet CIs pass.
624 bytes = 8 (~L2 per L1 blocks) * 78 bytes (empty block with just Deposited tx)
It's a delicate dance.

Propoer fix is to observe channel timeouts and sequencer window span so
that we can set an even larger target but still send txs quickly enough
when there's now tx volume.