Similarly, no need to implement retention in the first pass. It is highly S3-specific (because batch deletion on S3 is hard), and we may be rearchitecting it soon to be easier to port to other storage layers.