Making a scheduled job resumable after partial failure

A nightly job that dies at 80% should not start again from zero. How to make scheduled work resumable with a cursor, per-record idempotency, and a queue, and when to reach for the Workflow API instead.