(Premul)Rgba8::to_u32 now returns in little endian format.

[waywardmonkeys]

This matches the WebGPU spec which states that the internal layout of values is such that values in host-shared buffers are stored in little-endian format: https://www.w3.org/TR/WGSL/#internal-value-layout

[waywardmonkeys]

I will have to update the Vello shaders when this winds its way through Peniko and into Vello.

This also makes it the same as the as_u32 function in bevy_color which also returns a little endian value.

[raphlinus]

Agreed, I think it makes sense to standardize on this ordering. Thanks!

I am wondering about conversions into Rgba8, but I think that's less urgent. I'm also wondering (out loud) whether the From trait might be appropriate here.

[DJMcNab]

This raises questions about the bytemuck implementations. Endianness is like affinity, in that it's really hard to reason about quickly, but I think the bytemuck implementation is big endian? Should we re-order the fields to make that also be little endian?

We should have a test that they result in the same value, at least.

[waywardmonkeys]

I would be more inclined to remove that particular bytemuck impl in favor of using the new to_u8__array and the other stdlib features.

[DJMcNab]

Maybe? Certainly that would be reasonable for Rgb8, but I think for Rgba8 being able to cast slices would be useful (maybe we should give it an align of 4 as well?)