Symbolic Regression for PDEs using Pruned Differentiable Programs