[VE] Set getExtendForAtomicOps to ISD::ANY_EXTEND

[kaz7]

Summary

The implementation of subword atomics does not actually
guarantee the result is zero-extended, which now caused
failures after https://reviews.llvm.org/D101342 was landed.

[kaz7]

This fixes a problem caused in atomic_load.ll regression test.

llvm/lib/Target/VE/VEISelLowering.h

bool shouldInsertFencesForAtomic(const Instruction *I) const override {
     return true;
   }
   Instruction *emitLeadingFence(IRBuilderBase &Builder, Instruction *Inst,
                                 AtomicOrdering Ord) const override;
   Instruction *emitTrailingFence(IRBuilderBase &Builder, Instruction *Inst,
                                  AtomicOrdering Ord) const override;
   TargetLoweringBase::AtomicExpansionKind
   shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const override;
+  ISD::NodeType getExtendForAtomicOps() const override {
+    return ISD::ANY_EXTEND;
+  }
 
   /// Custom Lower {
   SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
   unsigned getJumpTableEncoding() const override;
   const MCExpr *LowerCustomJumpTableEntry(const MachineJumpTableInfo *MJTI,
                                           const MachineBasicBlock *MBB,
                                           unsigned Uid,
                                           MCContext &Ctx) const override;