Performance improvement

[israellot]

This small change reverses the origin loop order. By doing so landmarks become the first appearance of each hash, and collides are on forward ( left to right ) order.
This greatly improves performance, especially for larger inputs with small changes.
The original code, when matching a huge section would find the landmark, which would be further to the right, and cycle the collisions from left to right, getting a bigger match at every pass. With this modification most times the code can find the biggest match at the first pass, and skip checking if the other collisions are inside the already found match.

[endel]

Hi @israellot, thanks a lot for this pull-request! Sounds like you really dived deep on this!

I haven't been using this library lately though, can you confirm if the tests are still passing? Cheers!

[israellot]

Yes, the tests are still fine.
Thanks for the work you've done years ago, it has been useful for me on a data sync project.