How to Control LLM API Costs: Token Budgets, Batching and a Budget Guard in Python