So after running the output through xmlstarlet, seems as if there are five remaining issues:

• add xlmns attribute
• lang attribute missing in description elements
• ELI uses a country code of "ZZ" in one entry and "XN" in another that JOSM doesn't include as valid
• JOSM restricts max-zoom to 24 and ELI has one instance of 31 being used (which is naturally nonsense)
• there are a handful of complaints about a point element turning up unexpectedly (needs investigation)

* [ ]  there are a handful of complaints about a point element turning up unexpectedly (needs investigation)

This is due to JOSM limiting the number of points in a "shape" element to 999, potential solution: simply use a bounding box for these entries (8 as of right now), as we already have the bounding box, this means simply dropping the polygon in these cases.

For @grischard the following seem to have polygons with more than 999 vertices

Using BBOX for osmim-imagicode-usgs_250k_ant has 1222 vertices
Using BBOX for gsi.go.jp_airphoto_2020 has 1255 vertices
Using BBOX for bev-inspire-orthofoto has 1323 vertices
Using BBOX for South-Tyrol-DSM_2013 has 1013 vertices
Using BBOX for South-Tyrol-DTM_2013 has 1013 vertices
Using BBOX for USDA-NAIP has 1001 vertices
Using BBOX for san_antonio_river_2016_wms has 5509 vertices
Using BBOX for BDGEx_ctm_multi has 1001 vertices
Using BBOX for SP_Itu_-_Eixos has 2716 vertices

[...] the following seem to have polygons with more than 999 vertices
[...]
Using bounding box for SPW2021

There should be 121 vertices for SPW2021 (the only one I spot-checked).

[...] the following seem to have polygons with more than 999 vertices
[...]
Using bounding box for SPW2021

There should be 121 vertices for SPW2021 (the only one I spot-checked).

Sorry, that was a last minute rush before I went on vacation, I've corrected the list now.

I hope you had a good vacation. :)

Anyway, it looks like some of those high-vertices count polygons can/should be simplified. But probably not in this PR.
For example, san_antonio_river_2016_wms (5509) loses 3504 vertices with 0.5m max error simplification and 5091 with 1m max error.