Gated subspace alignment with drift compensation for parameter-efficient Class-Incremental Learning | Synapse