upgrade to wgpu-native v27.0.2.0

[Vipitis]

We now have a wgpu-native release for v27.0.2.0 skipping wgpu26
I should note that webgpu.h is very much out of sync and will likely require a lot more changes, there is also a PR for that: gfx-rs/wgpu-native#511

useful links:

• wgpu26 release
• wgpu27 release
• wgpu diff
• wgpu-native diff

Some working changes:

• DXC no longer requires dxil.dll, so the script in Add examples for instance extras #728 can be simplified (still no debug symbols :/)
• Polygon Mode Polygon mode support #693

todos:

• working examples
• passing tests
• wait for wgpu-native release https://github.com/gfx-rs/wgpu-native/releases/tag/v27.0.2.0
• use the issue template for this PR text, add links diffs
• multi-draw-indirect has been removed as a native only feature, supported if the downlevel_frag INDIRECT_EXECUTION is supported, which we can't check for.
• fix panic "Cannot remove a vacant resource" in the buffer destroy mem test, perhaps related to this change: https://github.com/gfx-rs/wgpu/blob/ee0719f2d84d99ad1a878c42afe8572dfc98276a/CHANGELOG.md?plain=1#L209
• check if depthSlice now works
• device.destroy() and query_set.destroy() might work now
• new instance extras regarding memory limits
• retest fatal errors and device lost errors
• [ ] can hardware raytracing be used now? No, it's an experimental feature which aren't enabled in wgpu-native 😞
• a lot of tests_mem fail now because there was a change to lifetimes before queue.submit()

[Vipitis]

Should be ready for review and working now.

The two items left open are device.destroy() and depth_slice_undefined which changed in the upstream diff and might need a proper implementation now.
I am not 100% sure what the device.destroy() method needs as the mem_tests currently pass... The query_set.destroy() is slightly different and was panicking. However I am not sure if my implementation and changes to the test is exactly what we should do.

[almarklein]

I had a look at the destroy methods. device.destroy() now indeed works! I tweaked the implementation of the queryset.destroy() and added a comment. We should be good for now.

As for depthSlice, fine if you want to look into that now. But can also be done later IMO.

[Vipitis]

I had a look at the destroy methods. device.destroy() now indeed works! I tweaked the implementation of the queryset.destroy() and added a comment. We should be good for now.

Thanks 🙏

As for depthSlice, fine if you want to look into that now. But can also be done later IMO.

I don't have a use for it myself, so would have to read up on the specifics. It's not crashing anything so should be fine to look into it whenever something comes up later.

[almarklein]

A new version of Pillow was released yesterday, which builds for pypy are targeted are 311. So I bumped our CI to use the matching pypy.

[almarklein]

Let's merge this and do a release.

[almarklein]

Thanks Jan! 🙏

[Vipitis]

Thanks for helping with the finish