Can Uncertainty Quantification Enable Better Learning-based Index Tuning?