Augmenting Bias Detection in LLMs Using Topological Data Analysis