Regroup collections by edition for minisearch results

[LionelZoubritzky-IGN]

The gpf_search_types result for "commune" in geocontext currently returns the following (when the limit is increased from default 10 to 20):

typename	score
BDTOPO_V3:commune_associee_ou_deleguee	128.00
BDTOPO_V3:commune	54.48
CADASTRALPARCELS.PARCELLAIRE_EXPRESS:commune	36.04
ADMINEXPRESS-COG.2025:chef_lieu_de_commune	35.47
BDTOPO_V3:zone_d_activite_ou_d_interet	35.41
ADMINEXPRESS-COG.2026:chef_lieu_de_commune	35.38
ADMINEXPRESS-COG.2017:commune	35.08
ADMINEXPRESS-COG.2018:commune	35.08
ADMINEXPRESS-COG.2019:commune	35.08
ADMINEXPRESS-COG.2020:commune	35.08
ADMINEXPRESS-COG.2021:commune	35.08
ADMINEXPRESS-COG.2022:commune	35.08
ADMINEXPRESS-COG.2023:commune	35.08
ADMINEXPRESS-COG.2024:commune	35.08
ADMINEXPRESS-COG.2025:commune	35.08
ADMINEXPRESS-COG.LATEST:chef_lieu_de_commune	34.62
ADMINEXPRESS-COG.LATEST:commune	34.57
BDTOPO_V3:zone_d_habitation	34.53
ADMINEXPRESS-COG.2026:commune	34.15
ADMINEXPRESS-COG.2025:commune_associee_ou_deleguee	33.64

It would be nice if gpf-schema-store could factorize the different available editions ("millésimes") of the same collections. This would allow ADMINEXPRESS-COG-CARTO-PE.LATEST:commune to appear possibly among the 10 first results. That would make the result clearer, and also more actionable for the LLM since ADMINEXPRESS-COG-CARTO-PE.LATEST:commune should be one of the most interesting collections for the LLM.

Also, the different editions that share the same score should appear with the most recent one first (the current order is 2017 -> 2025).

This issue is orthogonal to issue #71, which focuses on modifying the values of the scores (also useful in this example).