ASTRAL. Automated Safety Testing of Large Language Models