[llvm] [DA] Fix overflow of findBounds* in BanerjeeTest (PR #191232)

[Ryotaro Kasuga via llvm-commits]

kasuga-fj wrote:

> I will focus on adding regression tests to demonstrate if the Banerjee MIV test is necessary.

It's easy to craft such artificial examples. Or rather, they already exist, e.g., [Banerjee.ll](https://github.com/llvm/llvm-project/blob/c37764cc00f235bbd18e0ff496dfd2d40e6a64ec/llvm/test/Analysis/DependenceAnalysis/Banerjee.ll). I believe the important question here is how useful the Banerjee test actually is for real‑world cases. The [large-nested-6d.ll](https://github.com/llvm/llvm-project/blob/c37764cc00f235bbd18e0ff496dfd2d40e6a64ec/llvm/test/Transforms/LoopInterchange/large-nested-6d.ll) test would be one such case, but I have never seen any others. Also consider the implementation complexity and computational complexity of it. IMHO, to justify enabling the Banerjee test, we need to show more real-world cases where it can prove something meaningful.

https://github.com/llvm/llvm-project/pull/191232