[llvm] [JITLink] Some cleanups to EHFrameSupport (PR #66707)

[via llvm-commits]

lhames wrote:

The DWARF spec is handy for understanding eh-frames for exceptions but I don't think it constitutes a proper spec for them. E.g. the `CIE_pointer` in eh-frames is actually a delta _back_ to the CIE, rather than being an offset within the section as the docs claim. (And I think the idea of this being a link-time constant seems incompatible with `ld -r` in general).

The practical problem is just Darwin as far as I know: On some architectures (e..g arm64) we use relocations for some fields (pc-begin consistently, but others too). LLVM doesn't generate CIE edges (at least not these days), but I have seen them in the past, possibly from custom compilers. I'm inclined to pay the extra cost to check for and accept them where they're present.

https://github.com/llvm/llvm-project/pull/66707