Port union find to pytorch for the vertexing task
We previously used a DGL based implementation of the union find algorithm to go from track-pair wise compatibility to groups of compatible vertices.
We need to re-implement that algorithm in pure pytorch and add it to the onnx model so that in athena we get directly the vertex assignment, rather than the track-pair compatability. We can also look into optimising the union find algo to boost vertexing performance.