sokol_gfx.h resource bindings cleanup (issue #1037)

[floooh]

See: #1037

TODO:

• figure out why cgltf-sapp renders black, some uniform update confusion?
  • the uniform blocks for material and light params get confused, but funny enough, in the WebGPU inspector the uniform data looks correct!
  • metallic_params => binding=2 and light_param => binding=1 correct
  • metallic_params => binding=1 and light_param => binding=2 incorrect
  • resolved: this is actually correct WebGPU behaviour, even when weird. The dynamic offsets must be ordered by 'binding' not by BindGroupLayout entries.
• fix WebGPU dynamic offset order confusion
• rename sg_shader_desc.vertex_attrs back to attrs to be consistent with other sg_pipeline_desc
• consider removing the trailing _n from the new backend bind slots? no, it's good like that
• need to differentiate between max number of bindings, and the max binding index:
  • in the WebGPU validation layer:
    • need to fix bindslot collision funcs to use 2x uint64
    • fix backend specific out-of-range check ranges
  • max number of BG0 bindings is 8, but valid indices are 0..15
  • max number of BG1 bijndings is 40 (16 + 16 + 8), but valid indices are 0..127
  • in GL the storage buffer binding (max num storage buffers is 8, but valid indices are 0..15
  • size of wgpu bindgroups index needs to 128
  • size of GL storage buffer bind cache needs to be 16
• update backends:
  • Metal
  • D3D11
  • GL
  • WebGPU
  • dummy
• fix sokol tests
• fix raw samples
  • Metal
  • D3D11
  • GLFW
    • uniformarrays-glfw (needs sokol_debugtext.h)
    • vertexpulling-glfw
    • others
  • Emscripten
  • WebGPU
    • drawcallperf-wgpu (needs max bindings vs max bind slot fix in sokol_gfx.h!)
    • others
• update sokol-shdc:
  • auto-binding (manual @bindslot definitions required now)
  • create ATTR_* defines per program: ATTR_[prg]_[name]
  • rename SLOT_* defines to UB_[prg]_[name], IMG_[prg]_[name], SMP_[prg]_[name] and SBUF_[prg]_[name]
  • get rid of the 'merged reflection info', with the above per-program slot definitions that's no longer needed
  • manual bind slot definition via layout(binding=N)
  • check for bindslot collisions (different resources assigned to the same bindslot per linked program)
  • fix all sokol-shdc test shaders
  • update generators:
    • C
    • Zig
    • Odin
    • Nim
    • Rust
    • Jai
    • Yaml
  • update sokol-shdc docs
  • update sokol-shdc changelog
• util headers:
  • sokol_gfx_imgui.h
  • sokol_imgui.h
    • hlsl
    • others
  • sokol_gl.h
    • hlsl
    • others
  • sokol_nuklear.h
    • hlsl
    • others
  • sokol_debugtext.h
    • hlsl
    • others
  • sokol_spine.h
    • hlsl
    • others
  • sokol_fontstash.h
    • hlsl
    • others
• fix language bindings samples:
  • zig
  • odin
  • nim
  • rust
  • d
  • jai
• fix other projects
  • doom
  • v6502r
  • chips
  • qoiview
  • cimgui-starter-kit
  • ...
• test all samples on:
  • macOS + Metal
  • macOS + GL
  • iOS + Metal
  • iOS + GLES3
  • Windows + D3D11
  • Windows + GL
  • Linux + GL
  • Android + GLES3
  • WebGL2
  • WebGPU

NOTE: in order to allow combining 'random' vertex and fragment shaders in sokol-shdc, shader dialects with common bind space across stages (e.g. GLSL ssbos and WebGPU), sokol-shdc needs to assign vertex- and fragment-stage bindings to separate ranges. This may require some changes in the GL and WebGPU backends for the size of binding caches.

Q:

• Support glsl_location_n instead of glsl_name binding for uniforms? This would not be supported in GL 4.1 or WebGL2.
• Support 'sparse vertex attributes'? This would at least require wgsl_location_n and at least a couple of fixes where loops over the vertex attributes stopped on the first empty attribute => in sokol-shdc this is sort-of already supported because layout(location = n) works for vertex attributes