Use a more efficient datastructure for searching for duplicates

[ianwal]

Current searching requires comparing every video at least once, it takes n*(n-1)/2 comparisons.

There exists more appropriate data structures for searching for similar objects.

---

Vantage-point tree

A vantage-point tree would require rebuilding the tree when videos are added, O(n * log(n)), but searching would be greatly improved.

The distance function for the VP tree would be the similarity of a video. Currently, that is a number [0,100].

It could also be possible to store the individual PDQHashes in a VP tree, but the tree would become extremely large with large videos, and it would require storing a table for pdqhash:video. It would likely reduce search time though because it would avoid calculating the similarity of videos while traversing the tree.