Count BPE tokens in a text with tiktoken (o200k_base = GPT-4o default, or cl100k_base). An agent can check content fits an LLM context window before paying for an expensive call. Returns token count, char count and encoding. Honest scope: an OpenAI-style estimate — no open-source Claude tokenizer exists (exact Claude counts need the Anthropic API), Gemini differs too. Provided text only; nothing is fetched. Max 1 MB.