nx_self_repair_cycle_test.nx

nx_self_repair_cycle_test.nx -- the self-build cycle UPGRADED to self-REPAIR, drawing on the growable RESEARCHER knowledge base (nx_research_basis). Cardinal #25: when a proposal is bad, REWRITE it better -- don't just filter it. On UNSOUND, the loop: (1) diagnoses the failure (counterexample = WHY, CEGIS), (2) the RESEARCHER consults the basis for a modern, CITED alternative, (3) the VERIFIER re-proves the alternative 1:1, (4) ABSORBS the repaired capability -- and ESCALATES (hard no) ONLY when the basis has no alternative. Star case (real, verified both ways): `(mul x 2^k) == (shl x k)` is UNSOUND at the k>=W boundary; the basis's research-grounded fix is the k<W validity guard (egg POPL21; Hacker's Delight). Guarded, it is sound for all k<W, all x. Known answer: 2 unsound proposals -> 1 REPAIRED+ABSORBED + 1 ESCALATED. exit 0.