[lld] [lld][AMDGPU] Support R_AMDGPU_ABS32_(LO|HI) relocations (PR #191550)

[Fangrui Song via llvm-commits]

MaskRay wrote:

> I am not ELF or linker expert but to me looks good.
>
> [rampitec](https://github.com/rampitec) approved these changes [4 hours ago](https://github.com/llvm/llvm-project/pull/191550#pullrequestreview-4092798634)
>
> @[jhuber6](https://github.com/jhuber6) jhuber6 merged commit [3a47349](https://github.com/llvm/llvm-project/commit/3a47349c084217d83f9968db8b551a8526de9702) into llvm:main [22 minutes ago](https://github.com/llvm/llvm-project/pull/191550#event-24401706883)

This was merged with only one approval from a reviewer who self-described as not being an ELF/linker expert, while I was still listed as a requested reviewer. This felt like a pretty rushed merge as I do have comments:

* The only RELA-reachable call site is OutputSections.cpp, which reads back dynamic relocations to verify `writtenAddend == addend.`
* For new tests we prefer `$target-reloc-xxx.s` instead of `$target-xxx-reloc.s`

https://github.com/llvm/llvm-project/pull/191550