Generalising via Meta-Examples for Continual Learning in the Wild