Question about checkpoints in Sui Lutris

I wonder why we need checkpoints in Sui Lutris. What I understand about Sui Lutris is that it glues Fastpay with a consensus protocol (in this case, Narwhal and Bullshark), so we need a way to combine these two independent protocols, and in Sui Lutris, this is solved by using the consensus protocol to order transactions from the fastpath as well, but the transactions in the fastpath can be executed and ordered in parallel by the consensus engine so it can achieve low latency.

My question is, why do we still need a checkpoint? Well, if one wants to make a synchronous period to do a committee reconfiguration across epochs, one can just do one checkpoint at the end of an epoch, right? and there is no need to do a checkpoint for every n consensus commits at all (n = 1 in Sui Lutris). And what I understand is that the consensus already provides an order of transactions, so there is no need for a checkpoint to do it again, and they can use the order provided by the consensus, right?