Which less of an issue for log(n) scaling proposals as log(n) is eventually a constant, more of an issue for O(n) with a good initial constant!
On the other hand the proof costing weight is plenty of incentive for the decisions you mention favoring avoiding dust when they can.
For the subject of “your wallet might not implement stuff and abandon you” being functionally confiscatory. I think that’s a useful thing to consider, but I think it fundimnetally doesn’t get you far, as we’ve seen exactly that happen with stuff like “scanning blocks is unusably slow for our SPV wallet, guess we’ll just shut down”. The standard for confiscation has to be, I think, an assumption that the user can rip out the private key and use it with something that does work and that they’ll find it worth their time to do so.
Even bitcoin core has made it harder to use old wallets now, friction from software changing is just life.
As far as 7TB in storage costs that doesn’t seem that immodest to me, thinking in the long term. It’s a pretty good improvement on literally infinite. 