Improve async() by making its promises cancellable

[WyriHaximus]

Since async() returns a promise and those are normally cancelable, implementing this puts them in line with the rest of our ecosystem. As such the following example will throw a timeout exception from the canceled sleep() call.

$promise = async(static function (): int {
    echo 'a';
    await(sleep(2));
    echo 'b';

    return time();
})();

$promise->cancel();
await($promise);

This builds on top of #15, #18, #19, #26, #28, #30, and #32.

[azjezz]

I think this following case would fail 🤔

        $promise = async(function (): int {
            echo 'a';        
            await(async(function(): void {
              await(async(function(): void {
                await(sleep(2));              
              })());
            })());
            echo 'b';

            return time();
        })();

[SimonFrings]

@WyriHaximus looks good to me, only thing I just noticed is that the description for cancellation is missing inside the functions.php.

[WyriHaximus]

@WyriHaximus looks good to me, only thing I just noticed is that the description for cancellation is missing inside the functions.php.

@SimonFrings done 👍

[clue]

@WyriHaximus Some great progress in here, love this feature and how beautifully it aligns with our existing cancellation logic (#9 / #13)!

Added a number of smaller remarks below, otherwise looks like an excellent addition! Semantics make a lot of sense to me, so I'd love to get this in as is, but I do see some room for follow-up discussions once we address reactphp/promise#56 in the future 👍

[WyriHaximus]

@clue Yes this one is the missing link we need to make this package fully fit into our ecosystem. As mentioned on some of your comments, I will create a follow up PR to utilize WeakMap and WeakReference where possible to put the garbage collector in charge of cleaning up promise <-> fiber references. But for the sake of landing this feature, and requiring more time to fully test the behavior there I didn't update this PR with that.

[clue]

@WyriHaximus Some good progress! I agree the WeakMap implementation can be left up to a future PR, only added a couple of remarks below, otherwise LGTM!

[clue]

@WyriHaximus Thanks for the update, changes LGTM, so let's get this shipped!