Equivariant vs. Invariant Layers: A Comparison of Backbone and Pooling for Point Cloud Classification