QLESS: A Quantized Approach for Data Valuation and Selection in Large Language Model Fine-Tuning