Geometry of Linear Neural Networks: Equivariance and Invariance under Permutation Groups