feat: initial project scaffold

pyproject.toml (Python 3.12, uv), codex/ package (config, db, models),
infra/ (docker-compose + schema), .env.example, .gitignore, README.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

infra/docker-compose.yml

@@ -0,0 +1,47 @@
+# Paper knowledge base: Provenance + Citation graph + Semantic search
+# A single Postgres instance (pgvector) covers all three layers.
+# GROBID parses references / structure from PDFs without an arXiv source.
+#
+# Quick start:
+#   docker compose up -d
+#   psql "${DATABASE_URL}"          # see .env.example for default value
+
+services:
+  db:
+    image: pgvector/pgvector:pg17       # official pgvector image
+    container_name: papers-db
+    environment:
+      POSTGRES_DB: papers
+      POSTGRES_USER: researcher
+      POSTGRES_PASSWORD: change_me      # change before production use
+    ports:
+      - "5432:5432"
+    volumes:
+      - pgdata:/var/lib/postgresql/data
+      # schema.sql is executed automatically on first start:
+      - ./schema.sql:/docker-entrypoint-initdb.d/01-schema.sql:ro
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U researcher -d papers"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+    restart: unless-stopped
+
+  grobid:
+    # CRF-only image is sufficient for reference extraction and is lighter than
+    # the full deeplearning variant. Check https://hub.docker.com/r/grobid/grobid
+    # for the latest 0.8.x tag before deploying.
+    image: grobid/grobid:0.8.2
+    container_name: papers-grobid
+    ports:
+      - "8070:8070"
+    restart: unless-stopped
+    # GROBID is RAM-hungry; uncomment to cap usage on constrained hardware
+    # (e.g. Nvidia Jetson):
+    # deploy:
+    #   resources:
+    #     limits:
+    #       memory: 4g
+
+volumes:
+  pgdata: