Catch issues with context window size when someone wants to evaluate a large PDF - perhaps skip LLM-based tests in that case or truncate the recommendations

[nonprofittechy]