Axiomatising Incomplete Preferences through Sets of Desirable Gambles