FedSkip: Combatting Statistical Heterogeneity with Federated Skip Aggregation