When too many query/ref pairs are reported, final_stats uses too much of memory due to its naive Python implementation

[karel-brinda]

Happens with eg 1M queries

This is the problematic part: https://github.com/karel-brinda/mof-search/blob/e79b0c842ed919f1787a3071a52065d2317c8f71/scripts/final_stats.py#L109

Probably should be possible optimize by that the output is sorted according to ref (so it's sufficient to keep just stats for the last ref in memory)