Mistral token counting/batching

[pbeart]

Hi, I'm interested in making a PR so that the embedding methods chunk data to ensure each request has less than the maximum (16384) required by Mistral. Unfortunately, there doesn't seem to be a lightweight way to count tokens for Mistral, such as tiktoken for OpenAI models - I guess most of the time it's going to be safe to use tiktoken with a safety margin, but is there a way to count tokens accurately without a heavy library?