What problem does this solve or what need does it fill?

Currently meshes and textures that don't need modification on the CPU (RenderAssetUsages == RENDER_WORLD) still need to be decoded to a CPU format, extracted, then unloaded. This incurs a system RAM cost, can be slow, and blocks the main world during extracted asset loads.

It'd be ideal if these assets would directly upload to the GPU as a part of the asset load, preferably on a thread that doesn't block the main world.

What solution would you like?

• Use a static OnceLock<Queue> and static OnceLock<Device> to provide easy access to wgpu resources in Image and Mesh asset loaders.
• Check if the RenderAssetUsages requires main world access.
• If it's only supposed to be used in the render world, allocate a buffer.
• Obtain a staging buffer via Queue::write_buffer_with.
• Pipe std::io::Read results into the provided buffer (with the appropriate decoding).

This should also make it easier in future cases where we may need to support DirectStorage/Resizable BAR uploads of assets.

This is a bit hacky, and relying on globals in the form of static OnceLock-ed variables, but may be reasonable until wgpu supports multiple queues.

What alternative(s) have you considered?

Continue with blocking extractions and copies.

Additional context

Related: gfx-rs/wgpu#3698