Beyond speculation: Measuring the growing presence of LLM-generated texts in multilingual disinformation