Issue 8: Test questions used were not representative

[sanjaigridsandguides]

Problem

• The five input questions are all philosophical/subjective ("Is a perfect society possible?", "If you could invent one superpower...", "Who is your favourite actor?").
• Every single one was marked answered with 1–2 iterations.

Impact

• The submission does not demonstrate the retry loop at all (no no_answer outcome, no exhausted-retry path).
• Reviewer cannot see whether the agent actually rejects bad answers, because none were ever rejected.

Required Fix

• Replace at least two questions with crisply factual ones where a wrong answer is obvious (e.g. "What is the capital of Australia?", "What does the Annotated type in Python do?").
• Include at least one question where the validator is expected to push back.