Persistent tag counts from redis to mysql should better than race conditions if the persistence is triggered at the same time. Previously 1 attempt would just die, now it will silently stop rather than cause errors.
Improve old trending tag summarization so it correctly breaks up it's workload across multiple background deferred requests.
Additionally, this prevents a case where it would continuously loop failing to prune rows due to a logic error. This error previously wasn't observable, as it would do most of the work in the one request due to a bug where the limit on the rows to check was being ignored.