This was lost in the transition to streams and will be necessary for systems that need to resume big copy processes that become interrupted.
Last time we did this it was big/complex and it's unclear it will hold up well for streams like serialtiles. Worth exploring this and at the same time alternatives for big jobs like splitting them up into smaller independent bits (e.g. by bbox/zoom level) that can be parallelized/tracked on their own.
cc @GretaCB