posting same URL twice returns the same ids and deduped=true on the replay

Prove that workspace state committed through the op catalog survives every client-side flush event we care about: a hard reload, a fresh browser context with no IndexedDB, a cold load against an edge-cached snapshot, the cross-sub-doc batch path, and the job-create contract that downstream features depend on. If any of these break, the realtime + process-request story does not hold.