Touch test all subgroup builtins in vert, tesc, tese, geom, frag, compute shaders
before and after enabling extensions to make sure the correct errors are generated
and then not generated (after enablement).

Allow runtime-sized arrays of acceleration structures

Misc fixes from dgkoch

Don't prefix with "Capability" since the majority of them don't.
Also add missing CapabilityImageFootprintNV

Fixed .dll install on MSVC.

Build.gn: allow optimization in glslang lib and standalone

Update SPIRV-Tools, SPIRV-Headers

Headers update fixes validation of spv.shaderBallotAMD.

SPV 1.4: Add code generation for SPIR-V 1.4 features

...when doing a variable lookup on an array of constants.

This includes the test for the first feature of OpEntryPoint.

Latest known-good SPIRV-Tools: WARNING: Needs python 3.x.

You may need to install python 3.x and also
[windows users] may need to get it in your path in front of python 2.7.x.

If you don't need the external projects (i.e., SPIRV-Tools) and don't want
to upgrade to python 3, don't create the External subdirectory and CMake
won't require python 3.  (CMake has a bug that makes it find the wrong one
on a first pass if the top level does not require the version needed by
lower levels.)

For nonuniformEXT constructor, make a copy of the node to decorate

Add support for GL_EXT_buffer_reference2

Fix issues of explicit conversions.

If a shader's only use of float16 is for FConvert, add the Float16 capability

When visiting instructions, check that there are no 16-bit storage capabilities
and add the Float16 capability. Same for int8/int16.

For certain int8/int16/float16 conversions,they are conditionally allowed
when extensions defining explicit arithmetic types are enabled.

This was added some time ago, but I suspect largely unused.
Christoph pointed out of few contradictions to actual convention,
so that's fixed. But, I suspect it is largely incomplete.