feat: Token Plan 迁移与 .env 热加载,并修复日报 AI 摘要为空

Token Plan 迁移 / 配置热加载:
- configs/llm_models.yaml: 各场景切到 Token Plan(deepseek-v4.1-flash / qwen3.6-flash)
- 新增 configs/runtime_env.py: .env 按 (mtime_ns, size) 热加载并同步 os.environ,
  统一 env_get 取值;llm / embedding / vectorstore / mcp / pipeline 改用 env_get
- configs/loader.py / scripts/run_scheduler.py 等配套调整
- 新增 tests/test_hot_reload.py

日报 AI 摘要为空修复(2026-09-25):
- 根因: 推理模型的 reasoning token 与正文共用 max_tokens, 预算 1500 被"思考"
  占满 -> text_tokens=0 / finish_reason=length, 摘要静默为空且不重试
- daily_report 场景新增 max_tokens(默认 4000, YAML 保存即热生效);
  LLMConfig 支持可选 max_tokens; 分块预算 800 -> 2000
- _llm_call 拆出 _call_once, 正文为空时自动加倍预算重试(上限 16000),
  用尽才降级返回空串; 网络异常重试语义不变
- docs/user-guide.md 新增 FAQ; continuation.md 记录本次排查
- 已重跑 2026-09-25 日报(report_id=357)补回 466 字摘要

测试: 相关用例 56 passed(test_hot_reload 12 passed);
      ruff 无新增问题; 3 个 crawler 既有失败与本改动无关

tests/test_report_builder.py

@@ -132,8 +132,8 @@ class TestLlmCallRetry:
 
     def test_retry_then_success(self, monkeypatch) -> None:
         import scheduler.reporter as rep
-        monkeypatch.setattr(rep, "_LLM_RETRY_TIMES", 3)
-        monkeypatch.setattr(rep, "_LLM_RETRY_BACKOFF_SEC", 0.01)
+        monkeypatch.setattr(rep, "_llm_retry_times", lambda: 3)
+        monkeypatch.setattr(rep, "_llm_retry_backoff_sec", lambda: 0.01)
         client, n = self._fake_client(2)  # 前 2 次失败，第 3 次成功
         out = rep._llm_call(client, self._cfg(), "p")
         assert out == "今日要点摘要"
@@ -141,8 +141,8 @@ class TestLlmCallRetry:
 
     def test_exhausts_retries_raises(self, monkeypatch) -> None:
         import scheduler.reporter as rep
-        monkeypatch.setattr(rep, "_LLM_RETRY_TIMES", 2)
-        monkeypatch.setattr(rep, "_LLM_RETRY_BACKOFF_SEC", 0.01)
+        monkeypatch.setattr(rep, "_llm_retry_times", lambda: 2)
+        monkeypatch.setattr(rep, "_llm_retry_backoff_sec", lambda: 0.01)
         client, n = self._fake_client(99)  # 一直失败
         with pytest.raises(ConnectionError):
             rep._llm_call(client, self._cfg(), "p")
@@ -171,6 +171,84 @@ class TestLlmCallRetry:
         assert n["count"] == 3  # 2 块 + 1 次合并
 
 
+class TestReasoningBudgetEscalation:
+    """推理模型占满 max_tokens 导致正文为空时的预算升级（回归 9-25 无 AI 摘要）。"""
+
+    @staticmethod
+    def _empty_then_ok_client(empty_times: int):
+        """前 empty_times 次返回空正文 + finish_reason=length，之后返回正常摘要。"""
+        from types import SimpleNamespace
+
+        seen: list[int] = []
+
+        class Completions:
+            def create(self, **kwargs):
+                seen.append(kwargs.get("max_tokens"))
+                if len(seen) <= empty_times:
+                    return SimpleNamespace(
+                        choices=[SimpleNamespace(
+                            message=SimpleNamespace(content=""),
+                            finish_reason="length",
+                        )]
+                    )
+                return SimpleNamespace(
+                    choices=[SimpleNamespace(
+                        message=SimpleNamespace(content="恢复后的摘要"),
+                        finish_reason="stop",
+                    )]
+                )
+
+        return SimpleNamespace(chat=SimpleNamespace(completions=Completions())), seen
+
+    @staticmethod
+    def _cfg(max_tokens: int | None = None):
+        from llm.client import LLMConfig
+
+        return LLMConfig(
+            provider="qwen", model="deepseek-v4.1-flash",
+            api_key="sk-test", base_url="https://example.invalid/v1",
+            temperature=0.3, max_tokens=max_tokens,
+        )
+
+    def test_empty_content_escalates_and_recovers(self, monkeypatch) -> None:
+        """正文为空时自动加倍预算并最终拿到摘要（不再静默返回空）。"""
+        import scheduler.reporter as rep
+        monkeypatch.setattr(rep, "_llm_retry_times", lambda: 3)
+        monkeypatch.setattr(rep, "_llm_retry_backoff_sec", lambda: 0.01)
+        client, seen = self._empty_then_ok_client(1)
+        out = rep._llm_call(client, self._cfg(4000), "p")
+        assert out == "恢复后的摘要"
+        assert seen == [4000, 8000]  # 首次失败后预算翻倍
+
+    def test_scene_max_tokens_wins_over_default(self, monkeypatch) -> None:
+        """场景 max_tokens 生效；未配置时回退代码默认 4000。"""
+        import scheduler.reporter as rep
+        monkeypatch.setattr(rep, "_llm_retry_times", lambda: 1)
+        client, seen = self._empty_then_ok_client(0)
+        rep._llm_call(client, self._cfg(6000), "p")
+        assert seen == [6000]
+        client, seen = self._empty_then_ok_client(0)
+        rep._llm_call(client, self._cfg(), "p")
+        assert seen == [rep.DEFAULT_SUMMARY_MAX_TOKENS]
+
+    def test_all_empty_returns_blank_without_raising(self, monkeypatch) -> None:
+        """预算升级用尽仍为空时降级返回空串（日报仍可入库，不抛异常）。"""
+        import scheduler.reporter as rep
+        monkeypatch.setattr(rep, "_llm_retry_times", lambda: 3)
+        monkeypatch.setattr(rep, "_llm_retry_backoff_sec", lambda: 0.01)
+        client, seen = self._empty_then_ok_client(99)
+        out = rep._llm_call(client, self._cfg(4000), "p")
+        assert out == ""
+        # 4000 → 8000 → 16000（受 MAX_SUMMARY_MAX_TOKENS 上限约束）
+        assert seen == [4000, 8000, 16000]
+
+    def test_budget_never_exceeds_cap(self) -> None:
+        """升级预算不超过 MAX_SUMMARY_MAX_TOKENS，避免无限放大。"""
+        import scheduler.reporter as rep
+        assert rep.MAX_SUMMARY_MAX_TOKENS == 16000
+        assert rep.DEFAULT_SUMMARY_MAX_TOKENS > rep.DEFAULT_SUMMARY_CHUNK_MAX_TOKENS
+
+
 class TestCollectXwlb:
     """_collect_xwlb 取数逻辑：应查询日报前一日(已播出的联播)，并跳过内容提要。"""