Hypernetwork-based Meta-Learning for Low-Rank Physics-Informed Neural Networks