[LV] Correctly cost chains of replicating calls in legacy CM.

Check for scalarized calls in needsExtract to fix a divergence between
legacy and VPlan-based cost model.

The legacy cost model was missing a check for scalarized calls in
needsExtract, which meant if incorrectly assumed the result of a
scalarized call needs extracting.

Exposed by llvm#154617.

Fixes llvm#156091.

(cherry picked from commit 0aac227)

Author: fhahn

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -1152,7 +1152,10 @@ class LoopVectorizationCostModel {
   CallWideningDecision getCallWideningDecision(CallInst *CI,
                                                ElementCount VF) const {
     assert(!VF.isScalar() && "Expected vector VF");
-    return CallWideningDecisions.at({CI, VF});
+    auto I = CallWideningDecisions.find({CI, VF});
+    if (I == CallWideningDecisions.end())
+      return {CM_Unknown, nullptr, Intrinsic::not_intrinsic, std::nullopt, 0};
+    return I->second;
   }
 
   /// Return True if instruction \p I is an optimizable truncate whose operand
@@ -1665,7 +1668,9 @@ class LoopVectorizationCostModel {
     Instruction *I = dyn_cast<Instruction>(V);
     if (VF.isScalar() || !I || !TheLoop->contains(I) ||
         TheLoop->isLoopInvariant(I) ||
-        getWideningDecision(I, VF) == CM_Scalarize)
+        getWideningDecision(I, VF) == CM_Scalarize ||
+        (isa<CallInst>(I) &&
+         getCallWideningDecision(cast<CallInst>(I), VF).Kind == CM_Scalarize))
       return false;
 
     // Assume we can vectorize V (and hence we need extraction) if the