[llvm] Verify threadlocal_address constraints (PR #87841)

[Alex Voicu via llvm-commits]

AlexVlx wrote:

Hmm, that is an interesting question; whilst I recall having excised this from an actual app, it does look somewhat contrived / bogus looking at it with today's eyes. I think it's fine to `XFAIL` it so as to not be blocked on it and I will rework it in a future PR, since what it tests is relevant. At a broader scope, LangRef says `The first argument is a pointer, which refers to a thread local global.`, and it seems (to me) that the verifier check is a wee bit more strict, since otherwise a validly AS cast pointer would still meet the LangRef wording. It's unclear if that pattern could ever obtain naturally though, since the intrinsic is overloaded, so perhaps we just need to touch LangRef as well?

https://github.com/llvm/llvm-project/pull/87841