Size the visibility range binding for the buffer type it was given

[glashoff]

Objective

On WebGL2, a mesh with a crossfading VisibilityRange gets a pipeline that fails validation:

In Device::create_render_pipeline, label = 'pbr_opaque_mesh_pipeline'
  Error matching ShaderStages(VERTEX) shader requirements against the pipeline
    Shader global ResourceBinding { group: 0, binding: 14 } is not available in the pipeline layout
      Buffer structure size 1024, added to one element of an unbound array, if it's the last
      field, ended up greater than the given `min_binding_size`, which is 16

Fixes #21309.

Solution

Binding 14 is the visibility range buffer, and it is declared two ways depending on how many
storage buffers the device has. mesh_view_bindings.wesl:

const VISIBILITY_RANGE_UNIFORM_BUFFER_SIZE: u32 = 64u;
@if(AVAILABLE_STORAGE_BUFFER_BINDINGS__GE_6)
@group(0) @binding(14) var<storage> visibility_ranges: array<vec4<f32>>;
@else
@group(0) @binding(14) var<uniform> visibility_ranges: array<vec4<f32>, VISIBILITY_RANGE_UNIFORM_BUFFER_SIZE>;

layout_entries gave that binding Some(Vec4::min_size()) — 16 bytes — whichever branch had
been taken. That is right only for the storage declaration, where the array is runtime-sized and
one element genuinely is the minimum the shader can be said to require. The uniform declaration
is fixed at VISIBILITY_RANGE_UNIFORM_BUFFER_SIZE elements, so the shader requires all 1024
bytes, and a layout promising 16 is one a 16-byte buffer could satisfy.

Only the layout changes. The buffer bound there is already 1024 bytes, because
write_render_visibility_ranges pads it to exactly VISIBILITY_RANGE_UNIFORM_BUFFER_SIZE
elements on this path, so the larger min_binding_size is one it already satisfies.

The size now follows the binding type. VISIBILITY_RANGE_UNIFORM_BUFFER_SIZE was private to
bevy_render, so this exports it rather than writing the number out again here.

Testing

• The visibility_range example from Visibility Range example doesn't work in webgl2 #21309, built for wasm32-unknown-unknown with webgl2 at this commit and at its parent: it fails on the parent and runs here.

AI usage

The gap was found by Claude Opus 5, and the change and the verification runs above were drafted
and carried out with it.

I have read every line of this change and stand behind it.

[github-actions]

Welcome, new contributor!

Please make sure you've read our contributing guide, as well as our policy regarding AI usage, and we look forward to reviewing your pull request shortly ✨

[beicause]

Looks correct. I didn't tested it.

[beicause]

I think it's better to use checked_mul instead of saturating_mul, though they have no different here.