added visualization adapter for offline trace integration

[sherrylicodes]

This PR adds a visualization adapter layer for Issue #41.

It converts offline trace outputs into standardized visualization-ready data structures for the Streamlit UI. This includes flattening attention heads into residue-level connections and building a unified payload for visualization components.

This is intended to serve as the integration bridge between the reader/archive layer and the visualization UI.

[PranavNarala1]

I looked through the Streamlit upload and caching flow, and one thing I think would be worth checking is how the uploaded file is handled after it gets read for caching. Since the upload object is being used to build the cached reader, it might be safer to make sure the file state is reset afterward so it does not cause issues later in the session or on reruns. A small change like resetting the file pointer or using a less stateful way to access the uploaded contents could make this part more robust. I think adding a quick test for repeated uploads or reruns would also help confirm that the workflow behaves consistently.