Topology of diophantine sets: remarks on Mazur's conjectures