[llvm] [llubi] Add support for exposed provenance (PR #200596)

[Ralf Jung via llvm-commits]

RalfJung wrote:

> This matches Miri's behavior.

This is not correct. What you describe is in fact better than what Miri does. Miri does not "narrow down" the provenance of a wildcard pointer over time -- we just treat it like a wildcard on each access. We do this because the approach you use relies on the fact that on a load/store, there is at most one valid provenance for the access, which is true in llubi at the moment but not true in Miri (and will also stop being true in llubi when `noalias` or `nofree` or `captures` is added).

The semantics you describe are also known as [PNVI-ae-udi](https://www.open-std.org/jtc1/sc22/wg14/www/docs/n2364.pdf). 

https://github.com/llvm/llvm-project/pull/200596