It gets brought up because, at least from my perspective, none of the immediate mode toolkits are addressing this. I understand the desire to keep the code simple and only handle the rendering, but that is fundamentally opposed to the way some assistive technologies work. They need to access the UI as a state tree to even make sense. If the goal with these toolkits is to always avoid storing state information, that isn't ever going to work with screen readers. You can't push this task to the app developer, it really needs to be tied to the toolkit's internal hierarchy and focus handling.

My opinion for the last several years has been, if you use these toolkits, you might just have to accept that your app is not going to be accessible and is not really suitable for anything beyond highly specialized use.