Why is the tokenization for the japanese language different for query vs indexing time?

This a known issue currently. Tokenization  at indexing time and at query time is not the same and may lead to issues on occasion.