MT-SLVR: Multi-Task Self-Supervised Learning for Transformation In(Variant) Representations