fix(audit): P0 反思引擎 LLM 接入 + P1 hand run_id/skill triggers/pipeline v2

审计修复 Batch 1 (M4-02/M3-01/M5-01/M6-02):

P0 M4-02: reflection_reflect 从 KernelState 获取 LLM driver
  - 新增 kernel_state 参数，从 kernel.driver() 获取驱动
  - 自动路径(post_conversation_hook)已正常，手动 Tauri 命令路径已修复

P1 M3-01: hand_execute 返回 run_id 给前端
  - HandResult 新增 run_id 字段
  - execute_hand 结果包含 run_id.to_string()

P1 M5-01: skill-discovery 使用后端 triggers 字段
  - BackendSkillInfo 新增 triggers 字段
  - convertFromBackend 优先使用 triggers，fallback tags

P1 M6-02: pipeline_list 支持 v2 YAML 格式
  - scan_pipelines_with_paths 增加 v2 fallback 解析
  - 新增 pipeline_v2_to_info 转换函数
  - discovery.rs 导入 parse_pipeline_v2_yaml

注: M4-01 双数据库问题已在之前批次修复
     M6-01 route_intent 已确认注册，审计结论过时

desktop/src-tauri/src/intelligence/reflection.rs

@@ -760,9 +760,14 @@ pub async fn reflection_reflect(
     agent_id: String,
     memories: Vec<MemoryEntryForAnalysis>,
     state: tauri::State<'_, ReflectionEngineState>,
+    kernel_state: tauri::State<'_, crate::kernel_commands::KernelState>,
 ) -> Result<ReflectionResult, String> {
+    let driver = {
+        let kernel_lock = kernel_state.lock().await;
+        kernel_lock.as_ref().map(|k| k.driver())
+    };
     let mut engine = state.lock().await;
-    Ok(engine.reflect(&agent_id, &memories, None).await)
+    Ok(engine.reflect(&agent_id, &memories, driver).await)
 }
 
 /// Get reflection history