feat: V-learning架构 — 改Q→V

Q-learning问题: pass action Q值稳定偏高 → 永远pass
V-learning修复:
- 网络输出 14(Q-action)→1(V-state): P(win|state)
- NeuralBot: 对每个候选编码出牌后state → 选V值最高的
- pass和出牌都是候选, 公平竞争(不是固定action slot)

TrainingCollector:
- 新增 EncodeAfterPlay(view, tracker, afterHand, playedIds)
- EncodeState 支持 extraSeen 参数(模拟出牌时标记已见)

Python:
- train_v.py: V-learning训练器(load CSV→predict win/loss)
- network.py: OUTPUT_DIM=14→1

验证: 200局0错误, V-model loss 0.81→0.0001
需更多数据覆盖状态空间

PdkFriendServer/Logic/IsmctsBot.cs

@@ -392,6 +392,7 @@ namespace PdkFriendServer.Logic
                 int totalSims = allCandidates.Count * fastSims
                     + scoredCandidates.Count(c => c.winRate > 0.3) * (simsPerCandidate - fastSims);
                 Console.WriteLine($"[ISMCTS pos{view.MyPos}] {totalSims}sims => {desc}");
+                TrainingCollector.Record(view, _tracker, TrainingCollector.ActionToIdx(best.play), view.MyPos-1);
                 return best.play;
             }