Flamingo: A Visual Language Model for Few-Shot Learning