[wgpu-core] request_adapter fails when requesting with a specific backend

[rajveermalviya]

Description
Instance::request_adapter fails when instance is created with Backends::all(), and then a surface is initialized, and then request_adapter is called with a specific backend in an environment with 2 or more backends.

Repro steps

1. Ensure you have two or more different backends.
2. Update instance_request_adapter to pass a single backend like wgt::Backends::DX12/VULKAN/METAL instead of wgt::Backends::all().
3. Run the hello_triangle example
4. Notice request_adapter fails, if it doesn't fail then cycle through other available backends on Step 2.

Expected vs observed behavior
request_adapter should succeed

Platform
macOS 14, Metal + MoltenVK

[rajveermalviya]

Closing as seems to be dupe of #5324, which will be fixed by #5535

[Wumpf]

@rajveermalviya thanks for following up, did you confirm using the branch from #5535 that it fixes the issue for you? :)
(seems very likely indeed, but might be nice to have the extra datapoint :))

[rajveermalviya]

@Wumpf Yes, I confirmed that #5535 fixes the issue on at least MoltenVK + Metal setup (I haven't tested it on other platforms).
Thanks for working on it! :)

[Wumpf]

excellent, thank you!