[llvm] [SCCP] Extend `visitBinaryOperator` to overflowing binary ops (PR #84470)

[via llvm-commits]

XChy wrote:

> Some compile-time overhead, but probably acceptable: https://llvm-compile-time-tracker.com/compare.php?from=eb8f379567e8d014194faefe02ce92813e237afc&to=52f204492d88fad02ef9d510e23be3ceee63671c&stat=instructions:u Wonder whether there is any optimization potential in the ConstantRange implementation.

Out of curiosity, I'm wondering why such minor modification on SCCP brings obivous overhead on compile-time. From my perspective, this patch only replaces `add` with `addWithNoWrap`.

https://github.com/llvm/llvm-project/pull/84470