I think I can explain it: every offchain protocol can be done, no matter how complex it is, if you have N of N agreement. For the case of a bidi payment channel, you have exactly that : 2 of 2. updates (or "contract novation" to get all fancy) happen by agreement, and there's a trick to invalidate old contracts. A full L2 for 1M users could also do this, modulo computation and bandwidth. The reason nobody takes it seriously is because of the liveness requirement of N of N being unrealistic. We want for the bigger L2 systems to allow a decent level of passivity; with 2 of 2 you do have a liveness problem for sure but at least it's not "1 person out of 1000 wants to do a payment, an can't until literally all other 999 are online" :)