[wgsl-in] Implement quantizeToF16

[teoxoy]

Implement the quantizeToF16 built-in function from the WGSL spec.

[jimblandy]

The spec you want to follow is here.

The basic outline of this work is:

• At some point, add a snapshot test input file to naga/tests/in/math-functions.wgsl, and edit naga/tests/snapshots.rs to know about it.
• Add the function to naga::MathFunction in the Naga IR in naga/src/lib.rs. You'll get lots of compiler errors. Use todo!() to get things running again.
• Add the function to the WGSL front end, naga/src/front/wgsl. You'll figure it out.
• Extend the typifier in proc/typifier.rs.
• Extend the validator in valid/expression.rs.
• Add support to the WGSL backend in back/wgsl.
• Add support to the SPIR-V backend, using OpQuantizeToF16.
• Figure out what the corresponding functions are on Metal and HLSL, and add support to back/msl and back/hlsl.

You can use features like wgsl-in or spv-out to enable specific parts of Naga. By default, Naga doesn't enable any front- or backends, so you can pare down how much of the code base you want to deal with by just enabling the piece you're working on now.

[teoxoy]

You might find some more info in the issue that proposed adding this to the spec: gpuweb/gpuweb#704