Text this: Context-Aware Autoscaling for Cost-Efficient Large Language Model Inference With Prefix Cache Integration