Using diaries to collect data in resource-poor settings: questions on design and implementation