Quantifying Uncertainty and Variability in Machine Learning: Confidence Intervals for Quantiles in Performance Metric Distributions