Add 5 evaluation strategies with live demos

After showing that LLMs return unpredictable strings, teach 5 solutions:
1. Constrain answers (number/yes/no) — live demo
2. Multiple choice
3. Keyword matching
4. Semantic similarity (Token F1)
5. LLM-as-Judge — live demo with a judge LLM

Includes tradeoff comparison table.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: abossard