Is 'Right' Right? Enhancing Object Orientation Understanding in Multimodal Language Models through Egocentric Instruction Tuning