This just bring in new problems like people thinking they got paid but there being a refund trick.. This happens in the traditional system too, but at least when it exhibits total indifference to an account serving no purpose besides refund scams it is violating KYC principles instead of running as expected.
The point is that trade-offs are a fact of life. Some problems get solved, other problems appear. I'm not saying the new problems shouldn't be addressed, I'm just talking in general and I believe there are many kinds of solutions that can mitigate many kinds of problems. Just like in traditional banking. I just think it's pointless to try to invent and flesh them out here.
You could still decide what kind of use suits you the best. Regular people wouldn't (and shouldn't) need to know all the technicalities. A trusted party, maybe a bank, could provide their own integrated solution with whatever features they want to offer.
There are plenty of options in bank settlement protocols. I think the point under discussion is not banking improvement but bankless user sovereignty via technical means.
The smart contract writers sometimes fool themselves when working on the problem full time.. That's a bigger problem if the code is the contract instead of code attempts to honor the contract and a system with judgement can undo things that obviously fall bellow our ethical expectations like account ID swaps, supply chain attacks, kidnapping/intimidation and so on.