OGRUnionLayer: improve performance of GetFeature() once a full scan has been made

[rouault]

Dramatically improves performance of https://lists.osgeo.org/pipermail/gdal-dev/2025-October/061034.html

[jratike80]

Does this PR also make the documentation https://gdal.org/en/stable/user/ogr_sql_dialect.html#order-by outdated or does is only affect OGRVRTUnionLayer?

Note that ORDER BY clauses cause two passes through the feature set. One to build an in-memory table of field values corresponded with feature ids, and a second pass to fetch the features by feature id in the sorted order. For formats which cannot efficiently randomly read features by feature id this can be a very expensive operation.

[rouault]

Does this PR also make the documentation https://gdal.org/en/stable/user/ogr_sql_dialect.html#order-by outdated

no this is still relevant. The optimization in this PR helps for OGRVRTUnionLayer, but only if underlying sources have efficient random reading.