I think relearning all the mistakes of the past over again would be a set-back. Educational for sure, but quite a set-back. Probably better to just be more vigilant about removing old unused code from the existing kernel which they already do to some extent. The code and git logs already have a lot of comments that explain some of the lessons learned. I have not seen a need to rewrite it from scratch. With time parts of the kernel will probably get rewritten in Rust where it makes sense.

Microkernel, and hard RT scheduler from the start.

Stable ABI.