Reconstruction reducer currently aggregates all tabs into one text object; no tab‑level scope in KVS. Goal: introduce a document‑tab scope (doc_id + tab_id) and persist tab‑specific state.