Spatial Temporal Enhanced Contrastive and Pretext Learning for Skeleton-based Action Representation