On Enforcing Better Conditioned Meta-Learning for Rapid Few-Shot Adaptation