The truth of each channel still has to be a consensus. It doesn't change the fundamental issues of how to prove consensus about double-spends within the partition. Even there are chosen nodes who are signatories for determining the truth of the channel, this then not permissionless because the government can attack those specific nodes.
Detail your design and I will rip it to shreds.
You forget one critical thing which everyone forgets as soon as they read my consensus document.
The chosen nodes are ever changing in very short time windows (minutes) and are
not static. The fluidity of those nodes is critical to preventing exactly the problem that you are highlighting.
Furthermore the IP addresses of the chosen nodes are never known to the network so malicious agents, government or otherwise, are going to have a real hard time keeping abreast of not only
who the next signatories are to resolve consensus issues, but also
where to point their attack.
On top of that after a number of growth years there may be 100s or 1000s of signatories voting on consensus, and an attacker would have to take out at least 50% of them to cause even mild disruption. The worst case is transaction processing halts for a few minutes until the next set of nodes are eligible and the network continues operating. In the mean time an attacker has to identify who the next
m voters are, locate
n of them and take them out....rinse, repeat a few minutes later.