Maybe I didn't get the point. Of course we can't have exactly-once delivery directly in the layer of an unreliable network - but it seems pretty easy to construct it on a higher layer if your network stack supports at-least-once delivery: Just assign each message a unique ID during sending, then track those IDs in the receiver side and discard duplicates. And you need those IDs anyway so you can generate ACKs.
Isn't this basically what every "reliable" transport (TCP, HTTP3, message queues...) does?
Isn't this basically what every "reliable" transport (TCP, HTTP3, message queues...) does?