In-Place Tokenizer Expansion for Pre-trained LLMs