feat(phase-0): scaffold backend + web + docker + DB schema

- docker-compose.yml: timescaledb-ha (timescaledb 2.27 + vectorscale + pgvector + pgai)
  + backend (FastAPI, CUDA 12.1) + web (Next.js 14)
- docker-compose.gpu.yml: GPU profile overlay for RTX 3070 Ti
- build.bat: Windows bootstrap, auto-detects nvidia-smi and selects GPU/CPU compose
- backend: Dockerfile, pyproject.toml, FastAPI skeleton with /health and /health/db
- DB migration 001_init.sql: symbols (with trigram search), ohlcv_daily/1m (hypertables),
  macro_daily, trading_value_daily, news (vector embedding), predictions
  (with user_triggered flag for on-demand UX), prediction_outcomes, model_performance
- web: Next.js 14 + Tailwind + lightweight-charts placeholder
- README: KIS/DART/HuggingFace token issuance guides + 10 seed tickers + run instructions

backend/app/config.py

@@ -0,0 +1,24 @@
+from __future__ import annotations
+
+from pydantic_settings import BaseSettings, SettingsConfigDict
+
+
+class Settings(BaseSettings):
+    model_config = SettingsConfigDict(env_file=".env", env_file_encoding="utf-8", extra="ignore")
+
+    database_url: str = "postgresql+psycopg://stock:stockpw@db:5432/stock"
+    tz: str = "Asia/Seoul"
+    log_level: str = "INFO"
+
+    # 모델 디바이스 선택. 'auto'는 torch.cuda.is_available() 기반
+    model_device: str = "auto"
+
+    # External keys (옵션)
+    kis_app_key: str | None = None
+    kis_app_secret: str | None = None
+    kis_account_no: str | None = None
+    dart_api_key: str | None = None
+    huggingface_token: str | None = None
+
+
+settings = Settings()