[llvm] [llvm] Use computeConstantRange to improve llvm.objectsize computation (PR #114673)

[via llvm-commits]

serge-sans-paille wrote:

A potential fix is available there: #115504 interestingly, the underlying issue is that we cannot correctly represent large unsigned values *and* negative offsets and we need to make a choice.
It was working so far because those situations doesn't happen in the wild (who would statically allocate 2**32 bytes), but  `computeRange` make it more likely (a right shift is actually enough)

https://github.com/llvm/llvm-project/pull/114673