Storage Buffer Assets

[BrianWiz]

Challenge: Allow a custom material to read from the same storage buffer a compute shader writes to. #21

Adds a new example shared_storage. It runs a compute pass every frame and writes heightmap values to a storage buffer. A vertex shader reads from that and sets vertex heights (mesh is a plane). Result is an animated wavy terrain surface. On the CPU, it writes the "time" uniform every frame for animation purposes.

Adds a new function to the worker builder that allows you to retrieve a Handle<ShaderStorageBuffer> which can then be used on a custom material like so:

#[derive(Asset, AsBindGroup, Reflect, Debug, Clone)]
struct SharedStorageMaterial {
    #[storage(100, read_only)]
    heights: Handle<ShaderStorageBuffer>,
}

impl MaterialExtension for SharedStorageMaterial {
    fn vertex_shader() -> ShaderRef {
        "shaders/shared_storage_vertex.wgsl".into()
    }
}

You can retrieve the handle from the worker and clone it when creating the material.

fn spawn_mesh_system(
    mut commands: Commands,
    worker: Res<AppComputeWorker<SimpleSharedStorageComputeWorker>>,
    mut meshes: ResMut<Assets<Mesh>>,
    mut materials: ResMut<Assets<ExtendedMaterial<StandardMaterial, SharedStorageMaterial>>>,
) {
    if let Some(heights_handle) = worker.get_storage_buffer_asset_handle("heights") {
        let mesh = meshes.add(Plane3d::default().mesh().size(4.0, 4.0).subdivisions(10));

        let material = materials.add(ExtendedMaterial {
            base: StandardMaterial {
                base_color: Color::srgb(0.3, 0.7, 0.3),
                ..default()
            },
            extension: SharedStorageMaterial {
                heights: heights_handle.clone(),
            },
        });

        commands.spawn((
            Mesh3d(mesh),
            MeshMaterial3d(material),
            Transform::from_xyz(0.0, 0.0, 0.0),
        ));
    }
}

Notes:

• For the #[storage] attribute to work, it needs an asset handle like the code above. Bevy creates the buffer in the render subapp when the asset is first created. But, bevy_app_compute seems to have been designed to run on the main subapp. So I had to extract them back to the main subapp, but since the actual data is being stored on the GPU we're just cloning wgpu handles around anyway.
• The more I think about it, if we're going to have this feature I think all storage buffers should just use the same thing, I don't think it makes sense to leave it up to the user to decide how it's stored and accessed internally, it just adds an unnecessary burden to make a decision. So, I'll likely fix that before going out of draft mode.
• I think I added some code smell, so keeping in draft atm