We need to translate `might_contain` to a subfield filter in `SparkExprToSubfieldFilterParser` for probably more efficient filtering.