Bound large JSON data reads by payload behavior, not only row count

Large JSON-bearing rows can make a seemingly reasonable database row limit unsafe: query latency and payload size may cross service or statement timeouts long before the logical result-set limit is reached. When this applies, bound each upstream read to an empirically safe chunk, retain deterministic keyset pagination, and impose a separate total-scan ceiling. Preserve semantically required columns and fallback behavior; dropping valid fields or pushing down filters that bypass fallback semantics can turn a performance fix into silent data loss. This method is appropriate for large or highly variable row payloads; the safe chunk size should be measured for the actual projection and revisited when schemas or payload distributions change.