Upgrade to WGPU 22

[callym]

This currently doesn't pass the tests, I think because naga removed the PartialEq implementation on a bunch of types (gfx-rs/wgpu#5818)

[Elabajaba]

I haven't ran it locally to check if this is the issue, but the test output tends to change a bit on wgpu/naga upgrades. If it looks functionally the same it's probably fine to just overwrite the failing tests' expected outputs.

[teoxoy]

I put up gfx-rs/wgpu#6045 that will be backported and part of the 22.1 release.

[JMS55]

Naga 22.1 has released with PartialEq back on types.

[callym]

I've updated some failing tests - to me it looks like just the numbering in the error message is different

[Elabajaba]

I opened a PR for this with a few fixes and cleanups callym#1

We'll have to update bevy's capabilities detection as well.

[callym]

The test failures are assertions that were never hit before because they used #[cfg(debug)] instead of #[cfg(debug_assertions)]

[Elabajaba]

Oops, I guess I probably should've ran the tests locally when I fixed that.

[callym]

I don't think there's any way that check could ever work as the final shader string length is almost always going to be different to the initial string, for example here's one of the outputs:

initial shader string:
"#import test_module\n\nfn main() -> f32 {\n return test_module::entry_point();\n}\n"

final string after preprocess:
" \n\nfn main() -> f32 {\n return entry_pointX_naga_oil_mod_XORSXG5C7NVXWI5LMMUX();\n}\n"

[robtfm]

lgtm - haven't tested thoroughly but i agree the length check is vestigial, and the rest looks straightforward.