Improve CLAUDE.md documentation and make terminology agnostic

This change significantly expands CLAUDE.md with comprehensive project
documentation while making the codebase more framework-agnostic:

CLAUDE.md enhancements:
- Transform from minimal guidance file to comprehensive project guide
- Add detailed project overview with target audience and tech stack
- Document UV workspace structure with all three members (book, evals, scripts)
- Add complete prerequisites section (Dafny, Elan, UV, API keys)
- Document book structure with all 6 chapters
- Add detailed evals architecture with CLI usage examples
- Document DafnyBench implementation details
- Add deployment, architecture patterns, and troubleshooting sections
- Expand from ~80 lines to ~318 lines of structured documentation

Terminology improvements:
- Replace 'Lean agent' with 'MVP: the loop between LLM (◉) and a verifier (∀)'
  in book/00-index.md to avoid framework-specific language
- Update lean_agent_helix.py title to 'MVP a formal methods agent (it's a loop)'
- Change legend from 'LLM' to 'LLM API' for clarity
- Update figure caption to be verifier-agnostic
- Regenerate book/static/img/lean-agent-helix.png with new terminology

These changes make the project more accessible to formal verification
experts entering the AI/evals space while maintaining framework neutrality
in conceptual explanations.

Co-Authored-By: Claude <noreply@anthropic.com>

Author: quinn-dougherty