MoVEInt: Mixture of Variational Experts for Learning Human-Robot Interactions from Demonstrations