finalize_run() is the single source of truth for terminal run state.
Previously save_evaluation_results wrote a binary pass/fail status and
finished_at BEFORE rescue outcomes/metrics were saved, then finalize_run
overwrote it. Now scores only are written here; terminal state is set
once by finalize_run after all sub-results are persisted.
17-table PostgreSQL schema with full Chinese column comments,
FastAPI project structure (admin/common/modules),
DSL rule files, and schema migration scripts.
wren