Use VulkanShaderExamples rust-gpu port as difftests

[LegNeato]

While we have difftests, Use https://github.com/Rust-GPU/VulkanShaderExamples as benchmarks contains a lot of real-world shaders across multiple impls. We should use them as an additional difftest corpus. I am unsure if it is currently set up to render to a file to compare, but at least there is a "pause" method and perhaps a way to step through time.

related to #315

[LegNeato]

It will be built on top of: SaschaWillems/Vulkan#1220

[LegNeato]

Bummer, not going to be straightforward. There are differences for many shaders, because there is no frame stepping, and some (like the compute shaders) will basically never match because they run at full speed. Some animations do the same, where even using --paused or whatever in the benchmarks they still run. And of course there are float issues, but we can have a fuzzy threshold for those.

🟢 PIXEL-PERFECT IDENTICAL (14 examples)

No differences whatsoever between GLSL and Rust outputs:

• debugutils
• descriptorindexing
• displacement
• dynamicrendering
• inlineuniformblocks
• inputattachments
• multisampling
• negativeviewportheight
• occlusionquery
• shadowmapping
• shadowmappingomni
• specializationconstants
• texturearray
• triangle

🟢 VISUALLY IDENTICAL - Rounding Only (11 examples)

Single-bit differences (±1 value) that are imperceptible to human eyes:

• bloom - 1 pixel differs by 1 value
• sphericalenvmapping - 1 pixel differs
• texturecubemaparray - 1 pixel differs
• pipelines - 3 pixels differ
• textoverlay - 3 pixels differ
• texture - 7 pixels differ
• imgui - 4 pixels differ
• multiview - 10 pixels differ
• pbrbasic - 14 pixels differ
• texturemipmapgen - 10 pixels differ
• vertexattributes - 21 pixels differ

🟡 MINOR VISUAL DIFFERENCES (2 examples)

Small but potentially visible differences (max ≤5 values):

• deferred - 5 pixels, max difference of 3/255
• deferredmultisampling - 4 pixels with small differences

🔴 SIGNIFICANT VISUAL DIFFERENCES (32 examples)

Clearly visible differences in rendered output:

Compute Shaders:

• computeshader - 716K pixels affected (emboss filter)
• computeparticles - 2.3M pixels affected
• computenbody - 2.5M pixels affected (n-body simulation)
• computecloth - 224K pixels affected

Complex Lighting/Shading:

• ssao - 1.3M pixels, differences up to 51/255 (20%)
• hdr - 192K pixels, differences up to 47/255
• pbribl - 562 pixels, differences up to 23/255
• shadowmappingcascade - Only 79 pixels but up to 212/255 difference
  (83%!)

3D/Complex Rendering:

• gears - 330K pixels, differences up to 44/255 (17%)
• texture3d - 725K pixels affected
• terraintessellation - 1.4M pixels affected
• vulkanscene - 341K pixels affected

Others with significant differences:

• computecullandlod, descriptorsets, distancefieldfonts,
  dynamicuniformbuffer,
• gltfloading, gltfskinning, indirectdraw, instancing, multithreading,
• offscreen, parallaxmapping, particlesystem, pushconstants,
  pushdescriptors,
• radialblur, stencilbuffer, subpasses, tessellation, texturecubemap

[LegNeato]

Found this though: Rust-GPU/VulkanShaderExamples#5