[AMDGPU] Fix code object version not being set to 'none' (#135036)

Summary:
Previously, we removed the special handling for the code object version
global. I erroneously thought that this meant we cold get rid of this
weird `-Xclang` option. However, this also emits an LLVM IR module flag,
which will then cause linking issues.

libc/cmake/modules/LLVMLibCCompileOptionRules.cmake

@@ -215,6 +215,8 @@ function(_get_common_compile_options output_var flags)
       if(LIBC_CUDA_ROOT)
         list(APPEND compile_options "--cuda-path=${LIBC_CUDA_ROOT}")
       endif()
+    elseif(LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
+      list(APPEND compile_options "SHELL:-Xclang -mcode-object-version=none")
     endif()
   endif()
   set(${output_var} ${compile_options} PARENT_SCOPE)