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/reporter.py

@@ -17,7 +17,10 @@ import time
 from collections import Counter
 from datetime import date, datetime, timedelta
 from pathlib import Path
-from typing import Any
+from typing import TYPE_CHECKING, Any
+
+if TYPE_CHECKING:
+    from llm.client import LLMConfig
 
 from dotenv import load_dotenv
 from loguru import logger
@@ -519,10 +522,10 @@ def _generate_ai_summary(news: dict, cninfo: dict, day_str: str,
         return ""
 
     try:
-        from llm.client import load_llm_config, make_sync_client
-        config = load_llm_config()
+        from llm.client import SCENE_DAILY_REPORT, load_llm_config, make_sync_client
+        config = load_llm_config(scene=SCENE_DAILY_REPORT)
         client = make_sync_client(config)
-        return _llm_summarize(client, config.model, lines, day_str)
+        return _llm_summarize(client, config, lines, day_str)
     except Exception as e:
         logger.warning("AI 摘要生成失败: {}", e)
         return ""
@@ -548,8 +551,12 @@ def _split_lines_into_chunks(lines: list[str], max_chars: int = 3000) -> list[li
     return chunks
 
 
-def _llm_summarize(client, model: str, lines: list[str], day_str: str) -> str:
-    """LLM 摘要：单块直接总结，多块先分段总结再合并。"""
+def _llm_summarize(client, config: LLMConfig, lines: list[str], day_str: str) -> str:
+    """LLM 摘要：单块直接总结，多块先分段总结再合并。
+
+    config 为 llm.client.LLMConfig(daily_report 场景),提供 model / temperature。
+    """
+    model = config.model
     chunks = _split_lines_into_chunks(lines)
 
     if len(chunks) == 1:
@@ -609,9 +616,10 @@ def _build_prompt(lines: list[str], day_str: str) -> str:
 直接输出要点列表："""
 
 
-def _llm_call(client, model: str, prompt: str, max_tokens: int = 1500) -> str:
+def _llm_call(client, config: LLMConfig, prompt: str, max_tokens: int = 1500) -> str:
     """单次 LLM 调用（带重试），返回 strip 后的文本。
 
+    config 为 llm.client.LLMConfig(daily_report 场景),提供 model / temperature。
     失败按指数退避重试 `_LLM_RETRY_TIMES` 次（默认 3），全部失败则抛出最后一次异常。
     若 finish_reason 为 'length' 则说明达到 max_tokens 上限被截断。
     """
@@ -619,12 +627,12 @@ def _llm_call(client, model: str, prompt: str, max_tokens: int = 1500) -> str:
     for attempt in range(_LLM_RETRY_TIMES):
         try:
             resp = client.chat.completions.create(
-                model=model,
+                model=config.model,
                 messages=[
                     {"role": "system", "content": "你是 A 股日报撰写助手,输出简洁、有洞察的新闻摘要。"},
                     {"role": "user", "content": prompt},
                 ],
-                temperature=0.3,
+                temperature=config.temperature,
                 max_tokens=max_tokens,
             )
             content = (resp.choices[0].message.content or "").strip()

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: