> as library writer one can't still be sure the library will properly work across all async runtimes.
If you have assumptions that the runtime can't deliver, your library won't be useful with that runtime. True. Most obviously if you require async to mean "It just magically happens on another thread", and that's not what this runtime offers, too bad, that's not what you get.
It is conceivable that Rust will settle on Async always meaning Threads, and then this problem goes away, but I suspect it's far more likely (since Rust is adopted in embedded systems where nobody has any intention of "just" dropping in a complete threading system) that this persists, if you needed Threads then you'd better make sure you've actually got Threads, because duh, if there's only a single thread of execution then whenever you're not waiting for the async task to happen it isn't being worked on.
Latency has a similar issue, if you need latency guarantees you need a runtime with actual latency guarantees, Rust's async does not do that work for you.
But again, if down the road this is magically resolved (which I doubt) Rust can just ship a new edition with the new behaviour, which is the point you entirely ignored.
The problem is how to move forward, and I believe Editions are a significant improvement over the status quo for most languages in that regard. You can't expect to get everything correct in 1.0, and so you should already be planning for how to improve.
Currently in Rust async/await generators discussion has been put on hold as being a hot topic, and as library writer one can't still be sure the library will properly work across all async runtimes.
https://rust-lang.github.io/async-book/01_getting_started/03...
Same kind of 1.0 issues apply to Java, D, C#, C++,.... and they are specially bad in guest languages that take decisions on syntactic sugar for missing platform features, and when the platform adds those features, they usually end up with two ways of doing the same thing.