[llvm] Branch folding: Avoid infinite loop with indirect target blocks (PR #96956)

[Andy Kaylor via llvm-commits]

andykaylor wrote:

I hadn't noticed your question about whether x86 testing is sufficient. It probably isn't. I don't see anything target-specific in your test, and it looks like the branch folding pass is added for all targets if not explicitly disabled. Can you move the test to CodeGen/Generic?

https://github.com/llvm/llvm-project/pull/96956