[llvm] [AArch64][LV][SLP] Vectorizers use getFRemInstrCost for frem costs (PR #82488)

[Paul Walker via llvm-commits]

paulwalker-arm wrote:

If adding this new function that behaves differently to `getArithmeticInstrCost` for the reasons described is such a sticking point then I think we cannot safely use scalable vector `FRem` instructions on AArch64 and should instead change `LoopVectorize` and `SLPVectorize` to cost and emit the function calls directly.  Ideally I’d rather not complicate those passes but it would better reflect the transformation that is actually happening and thus remove my concerns and any need to modify TTI.

https://github.com/llvm/llvm-project/pull/82488