ACL: Adaptive Chunking of Large Language Models for Efficient Inference on Automotive Edge Devices