PWM: Policy Learning with Multi-Task World Models