Function chamfer 

pub fn chamfer(query: &[f32], doc: &[f32], dim: usize) -> f32

The Chamfer similarity of a query’s tokens against a document’s: for each query token, the best it does against any document token, averaged over the query.

This is the score the encoding approximates, and it is what to rerank the candidates with once the index has narrowed the collection down, the same way crate::Partitions::search reranks estimates against the full precision vectors. It is quadratic in the token counts, which is why it runs on the handful the index handed back and not on the collection.

Panics

If dim is zero, or if either side is not a whole number of dim dimensional vectors.