GLES explicit bindings

[kvark]

Every backend has specifics on how the bindings are treated. But all of them eventually need to ensure the naga::Module bindings are filled up, and this process is backend-independent. Now it's done in a separate function.
With all the improvements to GLES, it's now able to run the "particle" example on desktop:

RUSTFLAGS="--cfg gles" CARGO_TARGET_DIR=./target-gl cargo run --example particle

TODO: check how it works with Zed

[kvark]

Was debugging Zed on GLES, and it's quite tough. The tools for GLES on Linux are quite poor.
RenderDoc shows stuff that should be working, but the driver thinks otherwise.
Specifically, I'm seeing that, for each vertex in the quad, the clip distances generated from VS are roughly from 0 to 600. However, inside the pixel shaders they come out negative in the Y component specifically.

I'm believe I'm just seeing a driver bug here (on the AMD 6850U). Actually, a more reasonable explanation could be that I'm doing something wrong with the shared CPU-GPU memory. It behaves more deterministically under RenderDoc.

See clip distances

See the interpolated clip distances, normalized from (-500,500) range:

Notably, all of them are good except for Y component, which is below 0.5, which means negative.

It's behaving exactly as if the content_mask.size.x == 0.0