Allow WebGPU & WebGL in same wasm and detect WebGPU availability

[Wumpf]

Connections

• Based on Unify surface creation by introducing new SurfaceTarget enum #4984
• Fixes Dynamic WebGPU support detection #2804

Shoutout to @daxpedda who did a lot of the groundwork to make this possible!

Description
Allows to compile with both webgpu & webgl backends enabled. There's a new webgpu feature flag that is enabled by default for this purpose. When additionally enabling webgl this will no longer disable webgpu and both features can live side by side.
The WebGPU backend remains a special case in that far that an instance decides (via very simple support detection logic) upon creation if it is a WebGPU instance or a WgpuCore instance. I did a bunch of renamings in that area to make this more obvious. Also, all as_hal methods now return an Option (most already did) to make it possible for them to return None whenever both WebGPU & WebGL are enabled (only a WebGL/WgpuCore instance can do hal conversion). Similar patter applies to a variety of things.

DRAFT until #4984 lands
(makes the diff also significantly smaller)

Testing

• clippy etc.
• local web tests to make sure they still do webgl when asked to (workarounds were needed, see code comments)
• there's an egui branch that works with this https://github.com/emilk/egui/tree/wgpu-update

Checklist

• Run cargo fmt.
• Run cargo clippy. If applicable, add:
  • --target wasm32-unknown-unknown
  • --target wasm32-unknown-emscripten
• Run cargo xtask test to run tests.
• Add change to CHANGELOG.md. See simple instructions inside file.

[cwfitzgerald]

Need to see it rebased to fully review, but some initial comments

[cwfitzgerald]

Looks good outside of a issue with get_mapped_range

[cwfitzgerald]

Hell yes!

[cwfitzgerald]

@Wumpf please resolve conflicts, then merge

[Wumpf]

woo 🥳

[Keavon]

This looks very handy! Is there a documented list of things the WebGL version doesn't support? I assume compute shaders is one such obvious missing feature (although I wonder if there's been any effort put towards polyfilling compute shaders with fragment shader magic?).

[Wumpf]

@Keavon there's unfortunately no written out list since it's a bit hard to capture - some of the exact capabilities are determined at runtime. But you can check https://docs.rs/wgpu/latest/wgpu/struct.DownlevelFlags.html# for possible lacking features as well as https://docs.rs/wgpu/latest/wgpu/struct.Limits.html#method.downlevel_webgl2_defaults which lists out the reduced limits of webgl directly.
As you guessed there's no compute shaders and no effort to polyfill them - this would be huge undertaking and likely run into issues that would make you resort to cpu based emulation (most prominently there's no random access writes at all which bars the door to emulating the harder stuff like shared memory and don't even get me started about atomics ;))

[cwfitzgerald]

It's almost certainly easier to polyfill them in the user's application using application specific knowledge if you need them.