Why kafka broker didn't filter aborted transactions instead send them to consumers and let consuemers do the filtering?

From wiki， Note also the addition of the field for aborted transactions. This is used by the consumer in READ_COMMITTED mode to know where aborted transactions begin. This allows the consumer to discard the aborted transaction data without buffering until the associated marker is read. But why didn't broker just filter all the aborted record or return them as empty record to save the bandwidth and works of consumer developer?