Risk-sensitive Actor-Critic with Static Spectral Risk Measures for Online and Offline Reinforcement Learning