Genetic Quantization-Aware Approximation for Non-Linear Operations in Transformers