#38 WALK-FORWARD RE-OPTIMIZATION: enqueue a walk-forward run of ONE strategy. Body = the spec (inline strategy_spec or a saved strategy_id) + walkforward: {"folds": K (2..6), "optimize": [{"path": "entry.when.<...>", "grid": [...]}]} (≤3 tuned paths, ≤125 combinations — bigger grids are refused in plain words).

The engine splits the backtest window into K sequential folds; each fold f is graded OUT-OF-SAMPLE with the grid combination that had the best EV on folds 1..f-1 ONLY (fold 1 only ever trains). The result is the stitched OOS equity curve, the per-fold table {fold, window, chosen_params, is_ev, oos_ev} and the honest headline "walk-forward OOS EV" — every sub-window graded by run_backtest itself, never a second scoring path. Advisory like every engine read; the single-run verdict machinery is untouched.

Auth: same resolution as POST /v1/backtests (session cookie wins, else X-Vega-Key). Polled via the SAME GET /v1/backtests/{run_id}; the row's kind is 'walkforward'.