Projected Latent Distillation for Data-Agnostic Consolidation in Distributed Continual Learning