Derandomizing Multi-Distribution Learning