feat: 大模型使用场景化配置与去重多源记录

- 新增 configs/llm_models.yaml: 4 个场景(event_extraction/daily_report/stock_report/embedding)
  可独立配置 provider/model/api_key_env/base_url_env/temperature 等,含用途与模型要求说明
- 新增 configs/loader.py: YAML 场景加载器(优先级: CLI 参数 > YAML > .env > 内置默认)
- llm/client.py: load_llm_config 支持 scene 参数,LLMConfig 增加 max_attempts
- embedding/factory+remote+local: provider/model/batch_limit 支持场景覆盖
- scheduler/reporter+stock_reporter: 日报/个股摘要接入场景配置
- dedup: Fingerprint.source_ids 多源记录 + 旧库自动迁移 + DedupResult 多源字段
- scripts/run_dedup: uniques JSON 的 sources 字段 + data/deduped/{day}/sources.json 汇总
- scripts/run_event_extraction: 接入 event_extraction 场景
- 补充测试: 场景优先级/零值、多源合并、旧库迁移、embedding 场景覆盖

scheduler/stock_reporter.py

@@ -232,7 +232,7 @@ def _generate_ai_summary(company_name: str, announcements: list[dict],
                          news: list[dict], research: list[dict],
                          irm: list[dict]) -> str:
     """LLM 生成个股要点分析。"""
-    from llm.client import load_llm_config, make_sync_client
+    from llm.client import SCENE_STOCK_REPORT, load_llm_config, make_sync_client
 
     lines = []
 
@@ -277,12 +277,12 @@ def _generate_ai_summary(company_name: str, announcements: list[dict],
 直接输出要点列表:"""
 
     try:
-        config = load_llm_config()
+        config = load_llm_config(scene=SCENE_STOCK_REPORT)
         client = make_sync_client(config)
         resp = client.chat.completions.create(
             model=config.model,
             messages=[{"role": "user", "content": prompt}],
-            temperature=0.3, max_tokens=500,
+            temperature=config.temperature, max_tokens=500,
         )
         return (resp.choices[0].message.content or "").strip()
     except Exception as e: