Intersectional Bias in Japanese Large Language Models from a Contextualized Perspective