I think one reason most people wanting a next-gen-TeX project don't aim to compile to TeX is that some of the biggest pain points of TeX are baked into the core. So if you want to improve them, you need to change or replace at least some of the core layout algorithms, not just the front-end input language. For example a big wishlist item for many years has been some kind of improvement on TeX's quite frustrating figure placement, possibly with a more pluggable layout algorithm.