Count the LLM tokens in text for a given model or encoding — exact for OpenAI and open models (Llama, Mistral, Qwen, Gemma, DeepSeek, Command-R); labelled-approximate for Claude & Gemini.