[clang] [AMDGPU] Update f16 builtin definitions to use _Float16 instead of __fp16 (PR #182331)

[Shilei Tian via cfe-commits]

https://github.com/shiltian approved this pull request.

We need to change all of the remaining `__fp16`. For some reason those `__fp16` currently breaks some down stream workload, where HIP headers are using `_Float16` and the builtins are using `__fp16`, and for some reason they can't be implicitly converted.

https://github.com/llvm/llvm-project/pull/182331