Update hack to make Spark runner happy as well. (#31184)

Both input and materializationInput are the same logical PCollection,
but the Spark test is sensitive to which one is used.

sdks/java/core/src/main/java/org/apache/beam/sdk/transforms/View.java

@@ -298,7 +298,7 @@ private PCollectionView<List<T>> expandWithoutRandomAccess(PCollection<T> input)
               materializationInput,
               (TypeDescriptorSupplier<T>) inputCoder::getEncodedTypeDescriptor,
               materializationInput.getWindowingStrategy());
-      input.apply(CreatePCollectionView.of(view));
+      materializationInput.apply(CreatePCollectionView.of(view));
       return view;
     }