LoPT: Lossless Parallel Tokenization Acceleration for Long Context Inference of Large Language Model