cjk-aware-text-metrics

CJK/Latin weighted token estimation for multilingual LLM pipelines. Use when processing Japanese/Chinese/Korean text with fixed chars-per-token constants.