Hybrid actor-critic algorithm for quantum reinforcement learning at CERN beam lines