feat: split web and scraper into separate Docker images

- Dockerfile: replace single runner stage with web + scraper named targets
  - web: Next.js standalone only — no playwright, tsx, or scripts
  - scraper: scripts/lib/node_modules/playwright only — no Next.js output
- docker-compose.yml: each service pulls its dedicated image tag
- .gitea/workflows/deploy.yml: build both targets on push to main
- lib/db.ts: STALE_AFTER_MS reads PARK_HOURS_STALENESS_HOURS env var (default 72h)
- lib/park-meta.ts: COASTER_STALE_MS reads COASTER_STALENESS_HOURS env var (default 720h)

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

lib/db.ts

@@ -167,7 +167,8 @@ export function getMonthCalendar(
   return result;
 }
 
-const STALE_AFTER_MS = 72 * 60 * 60 * 1000; // 72 hours
+const STALE_AFTER_MS =
+  parseInt(process.env.PARK_HOURS_STALENESS_HOURS ?? "72", 10) * 60 * 60 * 1000;
 
 /**
  * Returns true when the scraper should skip this park+month.