Feature description
Maybe the implementation will not be a queue but it could be an error queue. Whenever we encounter a deserialization error (or maybe even any error) anywhere we could push the message in a go-through manner so that we can introduce a CLI or anything to recover those error messages (potentially by type).
It would greatly help in failure recovery / upgrade process where cases we didn't identify occur. We could provide a fix and launch the recovery process to process any message (maybe filtered) that couldn't be processed due to a bug during a given amount of time.