Multiple fixes to allow deployment to prod:

• API_URL env set for prod and dev docker compose frontend containers
• For prod, mount admin users list into API container
• Sort graph filter results by node name but link id
• Don't check admin user exists in internal db - need to fix checking of external user id in internal db
• Spin prod containers down before launching
• Separate run-prod and run-prod-fresh (if ssl cert to be regenerated) scripts
• Restore functionality where only skills with categories that are in current skill search are displayed initially (to reduce clutter in graph)