As discussed in D89381, we can use the range of an non-wrapping IV to
clamp down the max BTC in some cases.
This impacts a few other test cases that I still need to update and it
seems in some cases the new bound is more precise than the exact exit
count, which I still need to investigate. I just wanted to share this
early.
clang-tidy: warning: 'auto AR' can be declared as 'const auto *AR' [llvm-qualified-auto]
not useful