[llvm] [VPlan] Use DerivedIV in convertToStridedAccesses (PR #209611)

[Luke Lau via llvm-commits]

lukel97 wrote:

> Currently, integer Mul and PtrAdd without an underlying instruction aren't assigned any cost:

We currently cost the scalar sext/zexts, so the cost goes from 1 -> 0, which worsens the situation in https://github.com/llvm/llvm-project/pull/206896#pullrequestreview-4608064850 where we choose strides over segmented accesses because we don't account for the address offset computation

> The benefit of BinOpGEPFlagsTy is that we can use it in follow-ups to get GEP no-wrap flags in many places, which would be an improvement.

Is it possible to put up a stacked patch that shows where they would be used? 

https://github.com/llvm/llvm-project/pull/209611