[lld] [lld] Merge GOT entries for symbols that have been ICFed (PR #131630)

[Jessica Clarke via llvm-commits]

jrtc27 wrote:

> If both are same kind of symbols, I agree. But now I look closely at the pauth test failure, we need to handle function symbols and data symbols differently as what key we end up using in pauth depends on whether it's a func or not (IA vs DA).

Also STT_GNU_IFUNC vs STT_FUNC; calling the resolver directly as a normal function might be a bit of an unusual thing to do, but it's perfectly well-defined, and you wouldn't want it to accidentally call the returned function instead (or for IFUNCs to instead run the resolver as the implementation).

https://github.com/llvm/llvm-project/pull/131630