Staged Hybridisation for Visual Quantum Reinforcement Learning via Knowledge Distillation