Test the 10% energy and 5% runtime criteria

Test the 10% energy and 5% runtime criteria

Compute median energy and median runtime per variant from the 30 pairs, then test whether the candidate cuts median energy by at least 10% with no more than a 5% runtime regression. Report paired intervals. Deliverable: A scorer that prints each criterion as pass or fail with its interval and rejects any run whose output hash differs from the sorted reference.