[llvm] [SandboxIR][Bench] Test RAUW (PR #107440)

[Alina Sbirlea via llvm-commits]

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

LGTM with some notes.

For future patches, consider if you can generate IR that can be reused between benchmarks. For example, the IR in `generateRAUWIR` could be used instead of `generateBBWalkIR`. The concern here being future code explosion and keeping the focus on what's being tested.

You can also have one where you aggregate some of the API being tested in a single benchmark.

https://github.com/llvm/llvm-project/pull/107440