[BUG] Batched nn-descent isn't properly handling duplicates

[cjnolet]

The batched nn-descent is supposed to be returning a knn graph that contains only unique neighbor lists, however sometimes the distances vary slightly between batches for some overlapping neighbors, which causes duplicates to not be handled properly. We need to find a way to handle duplicates properly because it's causing invalid and incorrect results in the resulting knn graphs.