The Shape of Reasoning: Topological Analysis of Reasoning Traces in Large Language Models