Hm.

I would have just checked if the task is incomplete within parse_resume and then restart it there. It feels easier to understand and doesn't hardcode the connection issue triggered. I suppose the immediate issue is that if it's a valid RESUME then it might be resuming the pre-existing chunking process, though I forget the specifics for this case in particular.

I'm not sure if the ordering of resume dispatching vs this task erroring is reliable, or if it is currently, if it's easy to ensure it remains so, which is why I went with the done callback, as it will always be after the error has occurred if it is going to occur.

There's something still not right about this ordering though based on user testing. the approach from the resume side is probably more correct.