Use and limitations of learning curves for energy technology policy: A component-learning hypothesis