It's practical to restrict yourself to a sound subset of Scala (the soundness issue they mention applies to a specific usage pattern that you can ban with a linter), and the issue is outright fixed in Scala 3.
The undecidability is real but you at least get what you're paying for - a certain amount of type-level programming is practical and useful in Scala. I'm not sure it's the right tradeoff but it's on (or at least close to) the Pareto frontier in a way that many languages aren't.
But Scala is, according to that list, just as undecidable and unsound as TypeScript. Are the places where you can't trust your types just more esoteric there?