Harden scanned PDF OCR fallback

Fix document PDF text extraction

Fix: Remove non-existent frontend/public, app, pages dirs from Dockerfile COPY; use src/app structure

Fix: Install Node.js via official precompiled tarball in Stage 3 to resolve setup_20.x script error

Fix: Change npm ci to npm install to resolve missing Linux SWC binary on HF Spaces

Fix: Use node:20-slim and ignore TS/ESLint on build to fix HF Spaces build error

Deploy to HF Space