Matchlib returns unnecessarily many result pairs compared to the input strings

[matiaslindgren]

matcher.match_all_combinations accepts n strings and compares O(n^2/2) pairs, then returns results for all of those pairs (possibly filtered by a minimum_similarity parameter).
This makes the function generic, but also significantly increases the amount of results. When those pairs are processed by Radar, it searches for the pairs which maximize the similarity for each input string:
https://github.com/Aalto-LeTech/radar/blob/00a917ccb0f9a299aec265a2003776d9bee8ed8f/matcher/tasks.py#L120-L131

To significantly reduce the amount of returned results, matcher._match_all (or a new function) should also accept the current maximum similarity of each string, which allows the filtering to take place in matchlib, instead of Radar.