고속 엔진 undo stack 추가

docs/fast-engine-next-optimizations.md

@@ -5,12 +5,7 @@ serious traversal code should use the Cython `fast.pxd` API directly.
 
 Deferred work:
 
-1. Add an internal undo stack with `push_action()` / `pop_action()` so Python
-   callers can avoid tuple allocation when they need nested search.
-2. Keep traversal legal-action generation caller-buffer based. Consider a
-   reusable Python-wrapper action buffer only if wrapper profiling shows
-   `legal_actions()` allocation is material.
-3. Consider direct NumPy or feature-buffer output for RL pipelines instead of
+1. Consider direct NumPy or feature-buffer output for RL pipelines instead of
    building Python lists and converting later.
-4. Consider a single contiguous allocation for state arrays after profiling the
+2. Consider a single contiguous allocation for state arrays after profiling the
    simpler separate-allocation layout.