nx_p256_pointloop_bench.nx

nx_p256_pointloop_bench.nx -- reusable crypto perf workload for codegen A/B. A field-op-heavy P-256 point loop (double + add): long mem-bound limb chains, the real critical path where register residency (G1) pays off. Per-iter arena framing keeps memory constant. Used by bench/nx_codegen_ab.sh as a standing "racing crew" benchmark for any backend/codegen change (G1/G2/G3/...). expect_exit: deterministic (same under any correct codegen) -- A/B compares it.