AI startup Mistral has launched an API to moderate possibly toxic — or otherwise problematic — text in a range of languages.