Bridging Low-level Geometry to High-level Concepts in Visual Servoing of Robot Manipulation Task Using Event Knowledge Graphs and Vision-Language Models