[AMDGPU] Allow the `__builtin_flt_rounds` functions on AMDGPU (#90994)

Summary:
Previous patches added support for the LLVM rounding intrinsic
functions. This patch allows them to me emitted using the clang builtins
when targeting AMDGPU.