Implement async using workers

[AntonPieper]

I toyed around with Web Workers, Atomics, SharedArrayBuffers and OffscreenCanvas and this actually makes most async tasks possible without using asyncify! This does not require any dependencies or code transformations, it is just the browser's web APIs.

Changes

• Adds a raylib-wrapper.js file that is the new main entry point
• raylib.js is run on a worker (not called raylib-worker.js to make git happy)
• All necessary state is synchronized using SharedArrayBuffers
• sleeping is done on the Worker using Atomics#wait
• adds a server.py script due to security requirements for SharedArrayBuffer

---

Closes #45

[AntonPieper]

This is needed for GitHub pages, due to modern requirements for SharedArrayBuffer (after Spectre) and is a bit tacky by using a service worker that reloads the page with the needed headers set.

Ideally, GitHub pages would just set this header (like the the CORS headers), but that's not the case.

[AntonPieper]

Instead of modifying all the examples, I instead opted for backwards compatibility and added another example that does not use the raylib_js_set_entry function (straight from the official examples).

[AntonPieper]

Ideally this would be called raylib.js and the other file would be called raylib-worker.js (instead of raylib.js), but that would confuse git

[AntonPieper]

Because python -m http.server has no CLI options, this just starts http.server with the additional headers.

[AntonPieper]

Due to enable-threads.js, python3 -m http.server should work too, however this is more clean

[AntonPieper]

It can be used like this: ./server.py <optional_port>

[AntonPieper]

Relates to #22

[Azer0s]

Oh damn, I was working on the same thing, you beat me to it!
However, I would fully remove the backwards compatibility.

[AntonPieper]

The enable-threads.js hack makes this theoretically unnecessary, but if the hack is removed, this at least gives a message to the user.

[AntonPieper]

This is a little class I wrote to make working with the Data views of Linear memory a bit easier. You describe what attributes you store and their types and it generates the necessary data views, so you don't have to keep track of offsets yourself and instead can "just" refer to the variables created in the schema by name and it resolves to a typed array with the correct type at that location.