In a data processing environment, you are tasked with leveraging CDF to propagate deletes effectively. Describe how you would implement this in a Spark job, including the steps to ensure that deletes are correctly propagated across downstream systems. Provide a code snippet illustrating the key parts of this implementation.