Construct `OpAccessChains` for `RawBufferLoad` properly, or overhaul the API

[devshgraphicsprogramming]

struct Test {
    float4 mem1;
    float mem2;
    int mem3;
};

Glslang with GL_EXT_buffer_reference

%23 = OpAccessChain %_ptr_Uniform__ptr_PhysicalStorageBuffer_Test_r %_ %int_0
%24 = OpLoad %_ptr_PhysicalStorageBuffer_Test_r %23
%27 = OpAccessChain %_ptr_PhysicalStorageBuffer_float %24 %int_0 %int_1
%28 = OpLoad %float %27 Aligned 4

DXC with vk::RawBufferLoad<Test>(addr).mem2

%36 = OpBitcast %_ptr_PhysicalStorageBuffer_Test %35
%37 = OpLoad %Test %36 Aligned 4
%38 = OpCompositeExtract %float %37 1
    OpStore %30 %38

Now imagine I declared a large array for histogram or something as a member variable.

This is probably a dup of some issue (searched but couldn't find it in the ocean of open SPIR-V issues), since I'm certainly not the first person to discover this.

My suggestion to fix this is

namespace vk
{

namespace impl
{

// you'll need to magically generate and instantiate internally within Clang when you encounter it
template<typename T>
struct AccessChain;
/*
{
   // actually emit the OpLoad
   operator T() const;
   
   // all members redeclared but with types exchanged from U to AccessChain<U>
};
*/

}

template<typename T>
AccessChain<T> RawBufferLoad(uint64_t addr);
}

[natevm]

This appears to be a duplicate of #4986 and should be closed.