Benchmark detail

Every problem in the curated 100-suite, with its pass/fail status, step budget, wall time and token spend. Built from benchmark-results.json at deploy time — updated by prover bench --results.