GoViG: Goal-Conditioned Visual Navigation Instruction Generation via Multimodal Reasoning