HyperMARL: Adaptive Hypernetworks for Multi-Agent RL