v0.1.9.4 Walk-forward Evaluation

v0.1.9.4 ships the first walk-forward evaluation surface: calendar-time fold definitions, train-window scalar selection, selected-candidate test runs, walk-forward session identity, compact persistence, inspection helpers, degradation table output, and promotion-ready candidate extraction. Cost identity from v0.1.9.1, sweep retention from v0.1.9.2, and risk-chain identity from v0.1.9.3 are now wired into the walk-forward spine. Selection-integrity diagnostics, PBO/CSCV/CPCV, benchmark-relative metrics, all-candidate test retention, OMS/paper/live behavior, and ML-first tooling remain deliberately deferred.