The most obvious choice would be async/await support - no specific plans yet, though.
The concurrency model is, like C++ or even C#: sequential consistency for data-race-free programs (SC-DRF). No green threads or anything crazy, no 'message passing', just normal system threads, normal locks, you control how you access memory yourself. You do synchronization just like you'd do in C/C++.
If you go the route of async/await support - I went through the arduous process of figuring this out for zig (for which I think our design goals are nearly identical), and went through a couple different iterations. I have experience with using LLVM's coroutines API and abandoning it to codegen them manually. Would be happy to have a brainstorming chat with you sometime if you're interested.
I would appreciate it if more languages had something like ucontext_t or byuu/co, because I don't enjoy tagging every call site with "await" whenever I change a method to do something that might consume some time (and every call site of those functions containing call sites too, and so on), and because I do enjoy being able to write code that uses speculative execution (like for time-travel netcode for a 2d fighting game) without manually converting all my imperative code to a state machine like a compiler should do.
I am a bit of a snob, maybe, but I've come to think one of the bare minimum features for any language is copyable coroutines.
The concurrency model is, like C++ or even C#: sequential consistency for data-race-free programs (SC-DRF). No green threads or anything crazy, no 'message passing', just normal system threads, normal locks, you control how you access memory yourself. You do synchronization just like you'd do in C/C++.