feat(ai-service): replace mock AI/diff/quiz/defend routes with real service boundaries

[SharanyoBanerjee]

The AI service endpoints are currently mock implementations:

• generate.py checks ANTHROPIC_API_KEY but still returns mock text even when a key exists (apps/ai-service/app/routes/generate.py lines 18-28).
• quiz generation returns sample questions from query params rather than annotated code.
• defend evaluation passes after message count rather than evaluating the answer.
• diff returns a fixed sample diff instead of comparing the submitted code.

These are fine for scaffolding, but they do not yet implement the README's core Decode/Rebuild/Defend learning loop.

Suggested fix

Introduce clear service modules and request/response contracts:

• Claude client wrapper with timeout/error handling
• quiz generation from code/annotations
• defend question/evaluation flow
• actual diff/scoring engine
• tests for each route with mocked external providers

Acceptance criteria

• Routes no longer return fixed sample content in normal execution.
• Missing provider keys fail clearly or enter an explicit local mock mode.
• External AI calls are isolated behind service classes/functions.
• Each route has at least one test for success and one for error behavior.

---