fix: withhold AI training consent in robots.txt

[cblanc]

Summary

• Switch the robots.txt signal to Content-Signal: search=yes, ai-input=yes, ai-train=no: search indexing and AI grounding stay allowed, AI training is refused.
• Add Cloudflare's standard content-signals notice above the group. It defines the signals and states that a no is an express reservation of rights under Article 4 of EU Directive 2019/790.
• static/robots.txt and the committed build/robots.txt stay byte-identical.

Test plan

• pnpm build && PORT=8881 node dist/server.js
• curl -si localhost:8881/robots.txt: 200 text/plain, ai-train=no under User-agent: *

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[cblanc]

Not needed: postcodes.io keeps ai-train=yes (open data), as shipped in #1405.