We're seeing that some LLMs are using 65000+ tokens for raw text that is only 10-1000 characters long. This PR adds a max_token to be passed to the LLM API for each translation based on the length of the text.