feat(phase-4): LGBM 모델 + 앙상블 + 매칭/재학습 잡

- backend/app/models/lgbm.py: 종목 × horizon 별 LightGBM 회귀(y_ret_h)
  + 다중분류(y_dir_h, 3-class). joblib 으로 backend/data/models/{code}_h{H}_*.pkl
  저장. early_stopping(30). predict_one() 으로 최신 영업일 피처에 추론.
- backend/app/models/weights.py: ensemble_weights 테이블 IO,
  default w_chronos=0.6 / w_lgbm=0.4 (DB 행 없으면 fallback).
- backend/app/models/ensemble.py: Chronos sample 분포 + LGBM regression+cls
  결합. point/q10/q90 + prob_up/flat/down + direction 라벨. 한쪽 모델
  실패 시 다른 쪽 단독 fallback (cold start: chronos 단독).
- backend/app/pipelines/predict_one.py: predict_and_store(). 결과를
  predictions 테이블에 UPSERT, user_triggered 누적 OR. base_date = 마지막
  ohlcv 거래일, target_date = base_date + H 영업일(주말 스킵, 공휴일은
  매칭잡에서 자연 보정).
- backend/app/pipelines/match_outcomes.py: target_date == d 인
  user_triggered=TRUE 예측을 d 의 실제 종가와 매칭 → prediction_outcomes
  적재. direction_hit(±0.3% flat band) + abs_error. 실제 종가 없으면
  자연 skip.
- backend/app/pipelines/retrain_weekly.py: 시드 10종목 × H 재학습 +
  최근 30일 model_performance 적재.
- backend/app/db/migrations/003_ensemble_weights.sql: (code, horizon) →
  (w_chronos, w_lgbm, hit_rate_*, sample_count).
- backend/app/pipelines/scheduler.py:
    daily_batch    : 평일 16:00 KST
    match_outcomes : 평일 16:30 KST   ← 사용자가 확정한 매칭 시점
    retrain_weekly : 일요일 02:00 KST

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/app/models/weights.py

@@ -0,0 +1,75 @@
+"""ensemble_weights 테이블 IO. 기본 가중치 (chronos 0.6, lgbm 0.4)."""
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from sqlalchemy import text
+
+from app.db.connection import get_engine
+
+
+@dataclass
+class EnsembleWeights:
+    code: str
+    horizon: int
+    w_chronos: float
+    w_lgbm: float
+
+
+DEFAULT_W_CHRONOS = 0.6
+DEFAULT_W_LGBM = 0.4
+
+
+def load_weights(code: str, horizon: int) -> EnsembleWeights:
+    eng = get_engine()
+    with eng.connect() as conn:
+        row = conn.execute(
+            text(
+                "SELECT w_chronos, w_lgbm FROM ensemble_weights "
+                "WHERE code = :code AND horizon = :h"
+            ),
+            {"code": code, "h": horizon},
+        ).first()
+    if not row:
+        return EnsembleWeights(code, horizon, DEFAULT_W_CHRONOS, DEFAULT_W_LGBM)
+    return EnsembleWeights(code, horizon, float(row[0]), float(row[1]))
+
+
+def upsert_weights(
+    code: str,
+    horizon: int,
+    w_chronos: float,
+    w_lgbm: float,
+    *,
+    hit_rate_chronos: float | None = None,
+    hit_rate_lgbm: float | None = None,
+    sample_count: int | None = None,
+) -> None:
+    eng = get_engine()
+    with eng.begin() as conn:
+        conn.execute(
+            text(
+                """
+                INSERT INTO ensemble_weights
+                  (code, horizon, w_chronos, w_lgbm, hit_rate_chronos, hit_rate_lgbm, sample_count, updated_at)
+                VALUES
+                  (:code, :h, :wc, :wl, :hc, :hl, :n, NOW())
+                ON CONFLICT (code, horizon) DO UPDATE SET
+                  w_chronos = EXCLUDED.w_chronos,
+                  w_lgbm    = EXCLUDED.w_lgbm,
+                  hit_rate_chronos = EXCLUDED.hit_rate_chronos,
+                  hit_rate_lgbm    = EXCLUDED.hit_rate_lgbm,
+                  sample_count     = EXCLUDED.sample_count,
+                  updated_at = NOW()
+                """
+            ),
+            {
+                "code": code,
+                "h": horizon,
+                "wc": float(w_chronos),
+                "wl": float(w_lgbm),
+                "hc": hit_rate_chronos,
+                "hl": hit_rate_lgbm,
+                "n": sample_count,
+            },
+        )