Group Equivariant Video Action Recognition: Making Action-Recognition Networks Equivariant To Temporal Direction And Discrete Spatial Rotations