Can training neural language models on a curriculum with developmentally plausible data improve alignment with human reading behavior?