Show reasoning steps from o1 class models in italics in the dashboard

Related to: https://github.com/supercog-ai/agentic/issues/28 We should add a feature to the dashboard where it detects "reasoning" output from reasoning models like Claude and O3 and renders that text in a special way: maybe slightly-gray color plus italics. I am not sure if there's a standard way that Deepseek/Claude/o3 render their thinking steps or not. If it is standard we can just add dashboard support. If the output differs by each model then we should see if LiteLLM has some unifying abstraction that we can use.