refactor(engine): 移除 AutocorrectCut::has_generated 死字段

该字段在 autocorrect_cut_kbest 重建结果时赋值，作为公开结构体字段暴露在头
文件并配了契约注释，但 engine/、server/、windows/ 及全部测试都没有读取点。

静态优先的过滤实际发生在 autocorrect_cut_kbest 内部的 SearchHypothesis 上
（排序比较器 + 末尾按 front().has_generated 决定是否整批 erase），在 cut 重建
之前就完成了。头文件注释把契约归因到这个字段，读者会以为调用方依赖它。

删掉字段与赋值，契约说明改归真正的机制。SearchHypothesis::has_generated 是
活代码，保留。

engine/quanpin/quanpin_utils.cpp

@@ -980,7 +980,6 @@ std::vector<AutocorrectCut> autocorrect_cut_kbest(const std::string &pinyin, con
         AutocorrectCut cut;
         cut.edge_count = hypothesis.edge_count;
         cut.weight = hypothesis.weight;
-        cut.has_generated = hypothesis.has_generated;
         size_t position = length;
         const SearchHypothesis *current = &hypothesis;
         while (current->prev_index != kNoPredecessor)

engine/quanpin/quanpin_utils.h

@@ -112,11 +112,6 @@ struct AutocorrectCut
     // (neighbor gau -> gai, weight 13) regardless of dictionary frequency.
     size_t edge_count = 0;
     int weight = 0;
-    // True when any edge of this cut is an out-of-table generated pair. The
-    // search itself drops generated-containing cuts whenever a pure-static cut
-    // tops the ranking, so in-table inputs consume exactly the baseline cut
-    // set.
-    bool has_generated = false;
 
     bool empty() const
     {
@@ -152,8 +147,7 @@ Segments autocorrect_cut(const std::string &pinyin, unsigned autocorrect_types);
 // when the top cut is pure static all generated cuts are dropped before
 // returning -- in-table inputs get exactly the pre-generated-space cut set,
 // and out-of-table shapes surface only when no static cut explains the input.
-// Every
-// returned cut contains at least one corrected edge, so a fully legal input
+// Every returned cut contains at least one corrected edge, so a fully legal input
 // with no correction reading yields an empty vector (the caller owns the plain
 // segmentation). This is the query-time disambiguation surface of CN 101133411
 // B: ambiguity survives the cut layer and is settled by dictionary frequency.