[clang] [AMDGPU] Modify builtin def to take _Float16 for HIP/C++ and __fp16 for openCL (PR #166720)

[Shilei Tian via cfe-commits]

shiltian wrote:

Let's put this on hold. I still don't think the semantics of the new modifier is appropriate.

https://github.com/llvm/llvm-project/pull/166720