feat(digest): Phase 4 Global News Digest (cluster-level batch summarization)

7일 rolling window 뉴스를 country × topic 2-level로 묶어 매일 04:00 KST 배치 생성.
search 파이프라인 미사용. documents → clustering → cluster-level LLM summarization → digest.

핵심 결정:
- adaptive threshold (0.75/0.78/0.80) + EMA centroid (α=0.7) + time-decay (λ=ln(2)/3)
- min_articles=3, max_topics=10/country, top-5 MMR diversity, ai_summary[:300] truncate
- cluster-level LLM only, drop금지 fallback (topic_label="주요 뉴스 묶음" + top member ai_summary[:200])
- importance_score country별 0~1 normalize + raw_weight_sum 별도 보존, max(score, 0.01) floor
- per-call timeout 25s + pipeline hard cap 600s
- DELETE+INSERT idempotent (UNIQUE digest_date), AIClient._call_chat 직접 호출 (client.py 수정 없음)

신규:
- migrations/101_global_digests.sql (2테이블 정규화)
- app/models/digest.py (GlobalDigest + DigestTopic ORM)
- app/services/digest/{loader,clustering,selection,summarizer,pipeline}.py
- app/workers/digest_worker.py (PIPELINE_HARD_CAP + CLI 진입점)
- app/api/digest.py (/latest, ?date|country, /regenerate, inline Pydantic)
- app/prompts/digest_topic.txt (JSON-only + 절대 금지 블록)

main.py 4줄: import 2 + scheduler add_job 1 + include_router 1.
plan: ~/.claude/plans/quiet-herding-tome.md

app/services/digest/loader.py

@@ -0,0 +1,135 @@
+"""뉴스 7일 window 로드 + country 정규화
+
+- documents 테이블엔 country 컬럼이 없으므로 document_chunks.country 를 first non-null 로 조인.
+- chunk-level country 도 NULL 이면 news_sources.name prefix(ai_sub_group) 매칭으로 fallback.
+- 그래도 NULL 이면 drop(로그 경고).
+- ai_summary / embedding 이 NULL 이면 처음부터 제외 (재요약/재임베딩 0회 원칙).
+"""
+
+from collections import defaultdict
+from datetime import datetime
+from typing import Any
+
+import numpy as np
+from sqlalchemy import text
+
+from core.database import async_session
+from core.utils import setup_logger
+
+logger = setup_logger("digest_loader")
+
+
+_NEWS_WINDOW_SQL = text("""
+    SELECT
+        d.id,
+        d.title,
+        d.ai_summary,
+        d.embedding,
+        d.created_at,
+        d.edit_url,
+        d.ai_sub_group,
+        (
+            SELECT c.country
+            FROM document_chunks c
+            WHERE c.doc_id = d.id AND c.country IS NOT NULL
+            LIMIT 1
+        ) AS chunk_country
+    FROM documents d
+    WHERE d.source_channel = 'news'
+      AND d.deleted_at IS NULL
+      AND d.created_at >= :window_start
+      AND d.created_at < :window_end
+      AND d.embedding IS NOT NULL
+      AND d.ai_summary IS NOT NULL
+""")
+
+
+_SOURCE_COUNTRY_SQL = text("""
+    SELECT name, country FROM news_sources WHERE country IS NOT NULL
+""")
+
+
+def _to_numpy_embedding(raw: Any) -> np.ndarray | None:
+    """pgvector 컬럼을 numpy array(float32)로 정규화."""
+    if raw is None:
+        return None
+    arr = np.asarray(raw, dtype=np.float32)
+    if arr.size == 0:
+        return None
+    return arr
+
+
+async def _load_source_country_map(session) -> dict[str, str]:
+    """news_sources name → country 매핑.
+
+    name 은 '경향신문 문화' 형태이고 documents.ai_sub_group 은 '경향신문' (split[0]).
+    prefix 매칭이 가능하도록 첫 토큰 → country 로 인덱싱.
+    """
+    rows = await session.execute(_SOURCE_COUNTRY_SQL)
+    mapping: dict[str, str] = {}
+    for name, country in rows:
+        if not name or not country:
+            continue
+        prefix = name.split(" ")[0].strip()
+        if prefix and prefix not in mapping:
+            mapping[prefix] = country
+    return mapping
+
+
+async def load_news_window(
+    window_start: datetime,
+    window_end: datetime,
+) -> dict[str, list[dict]]:
+    """주어진 윈도우 안의 뉴스 documents 를 country 별 dict 로 반환.
+
+    Returns:
+        {"KR": [doc_dict, ...], "US": [...], ...}
+    """
+    docs_by_country: dict[str, list[dict]] = defaultdict(list)
+    null_country_count = 0
+    total = 0
+
+    async with async_session() as session:
+        source_country = await _load_source_country_map(session)
+
+        result = await session.execute(
+            _NEWS_WINDOW_SQL,
+            {"window_start": window_start, "window_end": window_end},
+        )
+        for row in result.mappings():
+            embedding = _to_numpy_embedding(row["embedding"])
+            if embedding is None:
+                continue
+
+            country = row["chunk_country"]
+            if not country:
+                # news_sources prefix fallback
+                ai_sub_group = (row["ai_sub_group"] or "").strip()
+                if ai_sub_group:
+                    country = source_country.get(ai_sub_group)
+            if not country:
+                null_country_count += 1
+                continue
+
+            country = country.upper()
+            docs_by_country[country].append({
+                "id": int(row["id"]),
+                "title": row["title"] or "",
+                "ai_summary": row["ai_summary"] or "",
+                "embedding": embedding,
+                "created_at": row["created_at"],
+                "edit_url": row["edit_url"] or "",
+                "ai_sub_group": row["ai_sub_group"] or "",
+            })
+            total += 1
+
+    if null_country_count:
+        logger.warning(
+            f"[loader] country 분류 실패로 drop된 문서 {null_country_count}건 "
+            f"(chunk_country + news_sources fallback 모두 실패)"
+        )
+    logger.info(
+        f"[loader] window {window_start.date()} ~ {window_end.date()} → "
+        f"{total}건 ({len(docs_by_country)}개 국가)"
+    )
+    return dict(docs_by_country)