Refactor backends

[almarklein]

In this PR:

• Introduce an auto backend, that we can later use to select the most appropriate backend.
• No more need to explicitly import the backend, because request_adapter() loads it automatically.
• Rename the rs backend to wgpu_native (closes Renaming rs backend to native #393).
• When a backend is loaded, it only replaces the wgpu.gpu object.
• The request_adapter() function is now a method of wgpu.gpu.
• Backend-specific logic is now only available from the backend's subpackage, as a functional API (enumerate_adapters(), request_device_tracing()).

Todo:

• Fix up codegen.
• Check download script.
• Move request_device_tracing()
• Update docs on needing to import the backend.
• Same for readme, examples, docstrings maybe.
• Merge Implement enumerate_adapters for rs-backend #412 in here.
• Make canvas arg in request_adapter() optional? -> No, this arg has been removed already in next webgpu spec.
• Check / review the pyinstaller hook (also wrt glfw).
• Deprecate wgpu.request_adapter() in favor of wgpu.gpu.request_adapter()?
• Update release notes.

[almarklein]

@Korijn what dou you think about this? Just import wgpu and get going.

[almarklein]

@Korijn when you review, can you have a look at the current pyinstaller hook?

[Korijn]

The hook looks good to me (and so does everything else!). This was a big one! 🐋