Add Vulkan Sampler Cache

[cwfitzgerald]

There is a similar limit to the limit we have on DX12 that there is a limit to the number of samplers that can be alive at once. On nvidia it's 4000 and is hit by the sampler tests in #6766. This ensures that there is a minimal amount of samplers by stealing a very similar cache structure to the DX12 work.

I have checked the mozilla tree and OrderedFloat 3.4 is vendored in there, so adding the dependency shouldn't be an auditing problem.

Checklist

• Run cargo fmt.
• Run taplo format.
• 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.

[jimblandy]

I think it's fine to take this PR, but it is too bad that we incur caching overhead and lose labeling even on GPUs that don't have any limit on samplers. Would it work to have the cache just sort of act as a pass-through when the limit is whatever ordinary GPUs return (i32::MAX??)?

Two other minor comments.

[jimblandy]

I think the pass-through idea could be implemented entirely locally to vulkan/sampler.rs. There'd just be a shortcut at the top of the create and destroy functions that checks the limit and skips the hash table altogether.

[jimblandy]

Here's a summary of the values GPUs have reported for maxSamplerAllocationCount:

https://www.vulkan.gpuinfo.org/displaydevicelimit.php?name=maxSamplerAllocationCount&platform=all

Value	        Reports
96	        77
1024	        110
2048	        42
4000	        19414
9999	        34
16384	        82
32768	        1686
65536	        7087
500000	        36
1000000	        4
1048576	        3463
1048580	        101
8388608	        10      
1073741824	944
4294967295	1451