Clean up WaveMask* implementations for SPIRV and GLSL

[fairywreath]

Slang has custom WaveMask* intrinsics which perform masked subgroup operations and are functionally the same as HLSL's WaveMultiPrefix* intrinsics. The WaveMask* functions are not properly implemented for SPIRV and GLSL however, SPV_NV_shader_subgroup_partitioned's ops should be used but instead regular non-masked subgroup ops are used. WaveMultiPrefix* support is already (correctly) implemented and the WaveMask* implementation for SPIRV and GLSL should be fixed.

[csyonghe]

We should just make sure WaveMultPrefix* is properly implemented for CUDA/SPIRV, and make WaveMask* alliases to those functions.