Integrating Symmetry into Differentiable Planning with Steerable Convolutions