Learning Interface Conditions in Domain Decomposition Solvers