Working with vulnerable groups in social research: dilemmas by default and design