[llvm] [llvm][amdgpu] Handle indirect refs to LDS GVs during LDS lowering (PR #124089)

[Jon Chesterfield via llvm-commits]

https://github.com/JonChesterfield approved this pull request.

Algorithm is as we discussed, implementation is completely clear, covers the problem described in the issue. Happy to accept!

I'm nervous about the extra callgraph walk but I think the cunning hacks to avoid it are now DoA given the transitive indirect calls. Something is going to have to crawl the whole graph after all. I should be able to follow this with a NFC that removes some of the redundant pointer chasing.

https://github.com/llvm/llvm-project/pull/124089