重构: 拆adversarial对抗评估(strategy核心库)

- 新建 df-ideas/adversarial_helpers.rs(366行): 6类型(EvaluatedBy/AdversarialEval/Argument/AnalystAnalysis/AssessmentLevel/Recommendation) + SYSTEM_PROMPT + 8纯fn(build_adversarial_prompt/parse_llm_eval/extract_json/parse_assessment_level/parse_recommendation/priority_label/assessment_desc/action_hint) + LlmEvalRaw中间结构
- adversarial.rs 930→596: AdversarialEngine impl保留 + pub use helpers(外部idea.rs路径零变更)
- lib.rs: mod adversarial_helpers(私有)
- ARC-260618-01-e(evaluate_with_llm一致性)逻辑保留(待决策, doc标注)
主代兜底(独立 -p df-ideas 避 df-ai): cargo 0 + test 20
strategy: 核心库, 纯函数/类型抽离, evaluate主入口+ARC-e保留
git add指定(df-ideas/*)

crates/df-ideas/src/lib.rs

@@ -1,6 +1,7 @@
 //! df-ideas: 想法池 — 捕获、评估、评分、晋升
 
 pub mod adversarial;
+mod adversarial_helpers;
 pub mod capture;
 pub mod promotion;
 pub mod scoring;