A Comparative Study on the Impact of Model Compression Techniques on Fairness in Language Models