This could result in a broken instance if it terminated in between the
two storage writes. Now we always persist the commit index before
applying the persisted entries and do so in a transaction with unstable
entries or snapshot if they are present.