WGSL back end can't translate GLSL `gl_WorkgroupSize` or SPIR-V `WorkgroupSize` builtin

[jimblandy]

As of gfx-rs/naga#2147, the WGSL back end will fail to generate code for modules that use naga::BuiltIn::WorkGroupSize, corresponding to GLSL gl_WorkgroupSize and the SPIR-V WorkgroupSize builtin.

It might make sense to remove Builtin::WorkGroupSize from Naga altogether:

• WGSL requires all compute shader entry points to have the a @workgroup_size attribute.
• Naga requires the workgroup size to be a static property of the entry point: the size is stored in EntryPoint::workgroup_size.
• The SPIR-V spec says that WorkgroupSize builtin is deprecated in favor of the LocalSizeId execution mode, which is a compile-time property of the entry point.
• GLSL requires the workgroup size to be specified statically as well via the local_size_x, local_size_y, and local_size_z layout qualifier identifiers; gl_WorkgroupSize is just a reflection of what those identifiers specify.

It might be possible to have the GLSL and SPIR-V front ends just replace uses of gl_WorkgroupSize and WorkgroupSize builtins with the given constants.

If we do keep naga::BuiltIn::WorkGroupSize, then we ought to generate code for it, probably by simply grabbing the value of EntryPoint::workgroup_size.

[jimblandy]

Oh, and per https://github.com/gpuweb/gpuweb/issues/1590:

"workgroup_size" builtin is not supported on some Apple platforms