We currently store publish states in a deleted state after an event process gets republished. Nothing then ever happens to that state. It actually means we have to be mindful of this when querying publish states too, and could be the reason for the failure of a migration test in https://jira.camunda.com/browse/OPT-5070. Over time, the more publish states that exist in a deleted state can have a performance impact on the index syncing background process.

We should either simply delete the publish state, or if we plan to use these to determine which version of an event process is the current, then be clearer about how the republishing handles the older publish states