Enhancing Symbolic Regression and Universal Physics-Informed Neural Networks with Dimensional Analysis