ES: convert all of Spain in one run, or per file

[ivorbosloper]

es reads 50 provincial GeoPackages (FEGA, 2025+) into one GeoDataFrame. #281 proposed a PerFileBaseConverter that converts file by file and merges the results, to keep memory bounded; it was closed, and its other fixes have landed since (the HCAT mapping in #327, areas of zero or less in #323).

• First run fiboa convert es on the full 2025 edition and note the peak memory.
• If that is fine, nothing to do: close this issue.
• If not, bring back the per-file conversion from PerFileBaseConverter: convert file by file, then merge #281 (its commits are still on the PR) as a small PR, using vecorel-cli's hilbert helpers instead of its own copies.

[m-mohr]

I think we should not do it, instead just create shortcut to create all separately with a merge shortcut. The same for Switzerland...

[ivorbosloper]

Measured on the full 2025 edition (50 provinces, 10.5 GB zipped, 26 GB extracted) on a 26 GB machine: after reading 46 of the 50 GeoPackages, the process footprint was 27 GB with swap growing, so I stopped it before the concat and write steps. One run does not fit; per-file conversion is needed.

vecorel-cli now has DuckDBBaseConverter.merge_parquet(paths, output_file, collection), which covers the merge half of #281.