Agreed. There's a lot of great index types out there - HNSW is incredible, but algorithms such as ScANN (and PQ) have their place in the ecosystem.
Tree-based vector indexes aren't bad either, especially if we can find a way to make the random projections more efficient.