How do you handle state management and recovery in a multi-stage document parsing pipeline if a worker node crashes mid-process at Reducto?
