This PR explores an improvement to the rapid search strategy, further cutting down the number of searches performed by keeping track of the maximum divergence for comparisons in each row of the distance matrix, rather than a single maximum divergence.
In practice this does seem to eliminate 10% or more of iterations through the inner loop. However, it doesn't seem to impact much on actual performance.
This PR explores an improvement to the rapid search strategy, further cutting down the number of searches performed by keeping track of the maximum divergence for comparisons in each row of the distance matrix, rather than a single maximum divergence.
In practice this does seem to eliminate 10% or more of iterations through the inner loop. However, it doesn't seem to impact much on actual performance.