MVP: Safeguarding guardrails and safe fallback responses

[andjump]

Context

See:

• docs/PRD.md §11 (Safeguarding rules)
• docs/architecture.md §5 (Guardrails)

The Talent Scout must be safe-by-design for young users.

Task

Implement guardrails to:

• refuse unsafe or out-of-scope requests
• prevent personal data exchange or private messaging
• provide approved fallback responses and signposting

Acceptance criteria

• Unsafe requests receive calm, supportive refusals
• Approved safeguarding copy is used consistently
• No raw sensitive content is stored by default

Out of scope

• Crisis intervention logic
• Replacing human safeguarding processes