[llvm] [LVI] Generalize mask not equal conditions handling (PR #92946)

[Antonio Frighetto via llvm-commits]

antoniofrighetto wrote:

I realize I've completely mistook inverse implementation. The original idea was to merge ranges `[SIGN_MIN, R.Lower)` and `[R.Upper, SIGN_MAX + 1)`, where `R` is the range obtained from EQ, then erroneously fell back to inverse; but even so, I now realize the original idea is wrong too. Is it still possible to save something out of it? Perhaps checking if we fall within `makeExactICmpRegion`?

https://github.com/llvm/llvm-project/pull/92946