[Clang][AMDGPU] Add __builtin_amdgcn_s_buffer_load_<T> (#203352)
Expose `llvm.amdgcn.s.buffer.load` intrinsic overloads for OpenCL via
`__builtin_amdgcn_s_buffer_load_*` builtins.
Function prototypes:
```
<return-type>
__builtin_amdgcn_s_buffer_load_<suffix>(
v4i32 rsrc, // 4-dword buffer resource in SGPRs
int offset, // byte offset from the base of the buffer
int aux); // cache-policy, control flags. Must be compile-time const.
```
Where:
```
return-type suffix
---------------- ------
char i8
unsigned char u8
short i16
unsigned short u16
int i32
half f16
float f32
v2i8 v2i8
v3i8 v3i8
v4i8 v4i8
v2i32 v2i32
v3i32 v3i32
v4i32 v4i32
v8i32 v8i32
v16i32 v16i32
v2f16 v2f16
v3f16 v3f16
v4f16 v4f16
v2f32 v2f32
v3f32 v3f32
v4f32 v4f32
v8f32 v8f32
v16f32 v16f32
```