Seeing Differently, Acting Similarly: Heterogeneously Observable Imitation Learning