[llvm] [LV] Don't ignore invariant stores when costing (PR #158682)

[Florian Hahn via llvm-commits]

https://github.com/fhahn approved this pull request.

> > Hmm, does that mean we now compute the cost of the scalar store for VF = 1, whereas before we would not compute it?
> 
> Yes, I think so.

Ok, that should be more accurate, as the scalar loop must execute the store on each iteration.

LGTM thanks.

https://github.com/llvm/llvm-project/pull/158682