It might just be familiarity but I think rayon.rs provides a nice model for how this could work. Rather than adding syntax, it basically provides a parallelized version of itertools.
This does require rewriting some code to take more of a functional style, but it’s a simple way to provide drop-in parallelism for people who want to try it.
I tested a small example of this in an early pep-703 thread and it provided the expected speedup on the nogil branch. Pretty nifty.