JIT: Move cpblk GC ref layout check back to lowering

The check in codegen runs only for arm64 but this check needs to happen
for arm32 as well. This moves the GC ref layout check back to lowering
and aligns it with xarch as well.

Fix dotnet#69976

src/coreclr/jit/codegenarmarch.cpp

@@ -2858,8 +2858,7 @@ void CodeGen::genCodeForCpBlkUnroll(GenTreeBlk* node)
     }
 #endif
 
-    if (!node->gtBlkOpGcUnsafe &&
-        ((srcOffsetAdjustment != 0) || (dstOffsetAdjustment != 0) || (node->GetLayout()->HasGCPtr())))
+    if (!node->gtBlkOpGcUnsafe && ((srcOffsetAdjustment != 0) || (dstOffsetAdjustment != 0)))
     {
         // If node is not already marked as non-interruptible, and if are about to generate code
         // that produce GC references in temporary registers not reported, then mark the block

src/coreclr/jit/lowerarmarch.cpp

@@ -493,10 +493,15 @@ void Lowering::LowerBlockStore(GenTreeBlk* blkNode)
 
         if (blkNode->OperIs(GT_STORE_OBJ))
         {
-            if (!blkNode->AsObj()->GetLayout()->HasGCPtr() || (isDstAddrLocal && (size <= copyBlockUnrollLimit)))
+            if (!blkNode->AsObj()->GetLayout()->HasGCPtr())
             {
                 blkNode->SetOper(GT_STORE_BLK);
             }
+            else if (isDstAddrLocal && (size <= copyBlockUnrollLimit))
+            {
+                blkNode->SetOper(GT_STORE_BLK);
+                blkNode->gtBlkOpGcUnsafe = true;
+            }
         }
 
         if (blkNode->OperIs(GT_STORE_OBJ))