Sanitize AI fallback messaging

[MyouzzZ]

Summary

• replace the fallback card's internal Gemini/API-key guidance with user-safe retry wording
• add an AI error message policy covering safe fallback behavior and details that should not be exposed
• note user-safe AI fallback messaging in the README feature list

Closes #14.

Validation

• .\.venv\Scripts\python.exe -m py_compile backend\main.py
• verified fallback copy via get_fallback_cards('Testing')[0]['content']
• git diff --check -- README.md backend/main.py docs/ai-error-message-policy.md