Combining school-catchment area models with geostatistical models for analysing school survey data from low-resource settings: Inferential benefits and limitations