Evaluating Tokenizer Performance of Large Language Models Across Official Indian Languages