[0011] Clarify behavior of SpirvType parameters and add SpirvOpaqueType

[cassiebeckley]

SpirvType now takes size and alignment as parameters, and a new template SpirvOpaqueType is added that does not. Also lets users specify whether a parameter for the type-declaration instruction should take an id or an immediate literal value.

Fixes #128.

[cassiebeckley]

@s-perron

[s-perron]

This is a pretty big change to the original proposal, but necessary to do what we need to do.

[llvm-beanz]

I think this LGTM. I wonder if there is some alignment we should aim for between this proposal and #154.

[cassiebeckley]

I'd considered using alignas, but it would require modifying the parser to have a different behavior than C++, which we were trying to avoid for the most part.

For example

typedef alignas(8) vk::SpirvType</* OpTypeInt */ 21, /* size */ 8, vk::ext_literal(8), vk::ext_literal(false)> uint8_t;

or

typedef vk::SpirvType</* OpTypeInt */ 21, /* size */ 8, vk::ext_literal(8), vk::ext_literal(false)> alignas(8) uint8_t;

would not be valid syntax.

[llvm-beanz]

Right... because alignas isn't valid for typedefs :(

That said, we might be able to implement this feature in terms of alignas in the future because this horrible thing is completely valid (with C23):

template<int Align, int Size>
struct alignas(Align) OpaqueType
{
  _BitInt(Size) V;
};