Document memory ordering guarantees for `std::thread::spawn()`

The documentation for std::thread::JoinHandle::join() guarantees that all memory operations performed by the joined thread are visible in the joining thread before join() returns:

[...] In other words, all operations performed by that thread happen before all operations that happen after join returns.

The documentation does not give such guarantees for std::thread::spawn() and the associated [Builder::spawn].

Obviously, such guarantee already exists implicitly:

fn main() {
    let s = "Foo".to_string();
    std::thread::spawn(move || {
        // All memory-operations performed by the spawning thread
        // are visible (happened-before) when the closure starts
        // executing, even though they have no explicit ordering.
        println!("{s}");
    }).join().unwrap();
}

Can we update the documentation on std::thread::spawn() (and friends) to explicitly make this guarantee?

Start with the linked documentation for std:🧵:spawn(), Builder::spawn, and JoinHandle::join(). Compare the existing memory-ordering language and document the corresponding guarantee for thread creation. Done means the spawn APIs explicitly state the ordering guarantee consistently with join().