audit-r2: structural no-op check, coerce args at boundary, latest-wins submit

audit: apply Codex pre-Phase-6 fixes (problem_id validation, no-op exact match, arg type coercion, forced-question classification)

phase-5 cleanup: episode_id in metadata, openenv push doc, README install line, psutil dev dep

phase-3: add operator_qa_log regression test, rename misleading restrict-tools test

phase-3: ShutdownGymEnvironment + server stack; fix audit.py to preserve scalar types in sanitize_args