For one applications need familiarity and uniformity. Games don't. For example games have exactly zero accessibility support, system actually knowing about what is being rendered is quite useful for it to provide tools like voice over. What about IME?
> If we’re going to optimize ONE stack
Why would we though? There are different classes of applications and there are multiple platforms with different conventions. Different use cases require different tools. Every time something "to rule them all" was tried, multiple niche tools grown around.