Changing PromiseExecutor to use a Rust VecDeque

[Pauan]

The behavior should be identical, but this dramatically improves the performance. In my animation test demo it went from ~10 FPS to ~60 FPS!

There are two reasons for this:

1. It maintains a VecDeque in Rust, rather than using the browser's event loop queue. This means much less interop between Rust -> JS -> Rust, which means less overhead.

2. It avoids creating garbage as much as possible.

   The old implementation created 3 functions and 2 Promises per event tick per Task. So if there were 100 Tasks that means it created 300 functions + 200 Promises on every event tick.

   The new implementation creates 1 Promise per event tick. So if there are 100 Tasks that means it creates 1 Promise on every event tick.

   That means multiple orders of magnitude less garbage.

It seems that avoiding garbage is much more important for performance than using VecDeque, but either way it's a huge performance benefit.

After these changes, the biggest performance bottlenecks seem to be from the to_js and to_js_string functions.

[Pauan]

Now that I got this done I can work on getting Discard landed.

[koute]

Thanks for the PR!

Hmm... this is certainly a creative way to implement a promise queue. Since this functionality is unstable anyway I'll merge it in as-is and we can continue fleshing it out. (We really need some tests and benchmarks.)