Anyway, we don't need a 100% effective ban to get an effect. A 90% ban may well be good enough for any practical purpose. Your example is a good one - sure, there are ways to get around 'internet kill switches' several countries have, but in practice these (unfortunately) work.
The fact that some people in a cafe can get one cryptocurrency transfer going is no defeat for a ban, so long as the ban is effective enough to reach its intended effects (e.g. making ransomware useless).
How would this be done? It only takes a single node capable of connecting to both networks to keep the whole thing working. There are already many nodes working with satellite connections so I'm pretty sure this can't be done even by state actors.
but the utility stays the same at $2.50 or $25,000 (offline-for-most, stateless monetary system with managed/predictable supply)
the same utility is inherited by most of other blockchain technologies
You can create the transaction object and hand that over. Transferrable literally as a file.
Instead of having over notes with the private key on them.
Eventually that transaction will need to be settled onchain. This can work in a world without a familiar looking ubiquitous internet. Regional internet cafes or even radio stations can settle signed transactions. People can have stocks of transactions and they just go to the cafe to settle those and also get their updated balance.
This capability always undermines a "ban crypto" thesis, or even a "crypto doesn't survive the apocalypse when the power goes out" thesis. It is merely a concept, just like computers communicating to each other is a replicable concept even if the head of state has an internet kill switch.