I wasn't able to run compute.toy on linux. Linux calls resizing immediately at the start of the program and invalidates current surface.

With this PR I'm adding support for surface resizing. Out of GPU memory case is not handled. In native I'll just stop the main loop, but I don't know what to do in web (perhaps nothing, because it wasn't handled before either).

[2023-05-07T17:54:16Z ERROR wgputoy] Unable to get framebuffer: The underlying surface has changed, and therefore the swap chain must be updated

I also changed a bit about how to choose the surface format. I give priority to Rgba8Unorm and Bgra8Unorm formats and store surface config instead of just surface format.

It also removes most of the warnings. From notable things, I changed device and instance creation. Instead of calling Default::default() each field is explicitly assigned to its value.

Bring this up, because it might be a non-trivial code transformation.

-                let workgroup_size = cap[1]
-                    .split(',')
-                    .map(|s| s.trim().parse().unwrap_or(1))
-                    .collect::<Vec<u32>>();
-                let workgroup_size = [
-                    workgroup_size.get(0).cloned().unwrap_or(1),
-                    workgroup_size.get(1).cloned().unwrap_or(1),
-                    workgroup_size.get(2).cloned().unwrap_or(1),
-                ];
+                // TODO: Handle error if failed to parse the capture
+                let mut sizes = cap[1].split(',').map(|s| s.trim().parse().unwrap_or(1));
+                let workgroup_size: [u32; 3] = std::array::from_fn(|_| sizes.next().unwrap_or(1));

Adds the following type aliases to the prelude.

alias bool2 = vec2<bool>;
alias bool3 = vec3<bool>;
alias bool4 = vec4<bool>;
alias float2x2 = mat2x2<f32>;
alias float2x3 = mat2x3<f32>;
alias float2x4 = mat2x4<f32>;
alias float3x2 = mat3x2<f32>;
alias float3x3 = mat3x3<f32>;
alias float3x4 = mat3x4<f32>;
alias float4x2 = mat4x2<f32>;
alias float4x3 = mat4x3<f32>;
alias float4x4 = mat4x4<f32>;

Since vecN<T> requires T to be a concrete scalar, and matNxM<T> requires T to be a float, this new set of type aliases covers all possible use cases (with the exception of f16).

The downside is that some shaders already contain a subset of these aliases, meaning some shaders will break and need to be fixed manually. Though this should be a 5-second fix in most cases, and the impact of such a change would be a lot greater further down the line.