Measuring interobserver variation in a pathology EQA scheme using weighted κ for multiple readers