Memory accounting designed off the hot path

Applies to: every managed-memory-accounted native operator (keyed aggregates, dedup, changelog normalize, joins, Top-N)

The accounting itself must not cost throughput. State footprint is tracked incrementally — only the groups a batch touches are re-measured, O(batch) rather than O(open state) — and there is no per-allocation JNI upcall into Flink's memory arbiter, the model Comet uses for Spark; the budget crosses JNI once at handle creation and is enforced by a local check. Measured cost: ~2% on the accounted keyed-tumbling bench, statistically unchanged on the unaccounted hot paths (66fcfe3, 2c1c487).

The GROUP BY aggregate's touched-group measurement runs twice per row (before and after the fold), and sizing its cached last-emitted tuple — the same cache described on Aggregate specialization fast paths — re-walked the tuple's ScalarValues each time. A 2026-07-12 q17 profile flagged the cost, so the size is now cached next to the tuple and maintained wherever the cache changes, making the per-row measurement pure arithmetic.