OPT-06: within-solve basis-reuse hull classification

Descriptive Release measurements for one implementation and three small 2D fixtures. No speedup or general scaling claim is made.

Measured source commit: ef70e8c580a9556b0f5d96e49070d33221bf5b64
Evidence: 42 measured rows, 0 failed rows; report status remains descriptive because no separate limiter profile was collected.
Timing scope: steady_clock around reduce_hull or reduce_hull_basis_reuse only; point cloud setup, HiGHS LP workspace creation, and independent hull-equivalence verification excluded.

Median durations

Per-method summary

case_id	method	attempts	successes	median_ns	min_ns	max_ns	Membership LP calls (excludes Phase-I solve)	Phase-I basis solves	median_pivots	median_degenerate_pivots	median_bulk_candidates	median_bulk_removals
degenerate	baseline	7	7	635900	622400	689200	7	0	0	0	0	3
degenerate	basis-reuse	7	7	51700	50200	53200	0	1	4	0	4	2
interior-basis	baseline	7	7	465600	445600	511700	5	0	0	0	0	1
interior-basis	basis-reuse	7	7	46500	45200	48600	0	1	4	2	2	0
permuted	baseline	7	7	922200	872700	1005200	10	0	0	0	0	6
permuted	basis-reuse	7	7	54600	50100	56100	0	1	4	0	7	2

Evidence and limits

Every warmup and timed pair returns complete reductions. For each timed pair, the harness retains the exact outputs from the timed invocations and, after timing stops, checks both hull containments using fresh registered HiGHS LP membership queries. This includes the interior-basis fixture: its baseline is complete, and the basis-reuse result reports structural basis IDs [0, 3, 4] and retains center ID 0. The focused unit test independently asserts the completed baseline, basis membership of ID 0, and both containments.

Hardware/compiler/build details and all raw rows are retained in records.json, trials.csv, summary.csv, raw stdout JSONL, raw stderr, and the sanitized build log. LP/pivot/bulk counters describe operations, not elapsed-time components.

The fixtures are intentionally small and do not model the fuel-consumption application or establish production workload scaling. Timed methods are alternated, but no tuned performance verdict is drawn; a dedicated limiter profile and broader workload family would be needed before claiming a winner.