[llvm] [LV] Implement integer part of VPDerivedIV cost model (PR #198252)

[David Sherwood via llvm-commits]

david-arm wrote:

The saxpy_5 loop also doesn't include the cost of the `Cost of 0 for VF 2: vp<%7> = SCALAR-STEPS vp<%6>, ir<5>, vp<%0>` recipe, which isn't necessarily free and I intend to fix in a follow-on PR. I think this test is extremely fragile because the cost for VF 2 is *very* close to the scalar loop cost. Any minor improvement or tweak in the vplan cost model will constantly throw move this below and above the threshold.

https://github.com/llvm/llvm-project/pull/198252