ReFS does block cloning, so it has a built in form of dedup. It's not real dedup but it will help, it basically makes synthetic full backups take no space.
However, while most people are now fine with ReFS, it's not completely solved, and you'll want a lot of memory in the repository server (at a guess for your data size I'd say 96-128GB).
You may also want to look at 2016 with NTFS and dedup. The latest patch supposedly fixed the final issues with dedup on 2016 so that opens up switching to ReFS later once it stabilizes fully without an OS reinstall.
Why do you say you'd have to reverse incremental? As far as I'm aware there isn't any significant space savings in doing so and the recommended is forward incremental