I can't say for certain that any .unl code info is not stored somewhere in RWFS, but not as far as i know.

Garmin devices certainly store both critical and non-critical information in the non-vol region (154 for your device). Some of that is temporary (can be cleared via the device settings in the Diagnostic Page menu). Some other info, specifically that which is unique to the device cannot be "cleared" as before, but it can be fully erased and that may also include other info gathered by the device (we don't know for sure). The absolutely unique critical info is also stored elsewhere on the device and it gets rewritten somehow to rgn154 when the device next reboots however other info maybe does not. In fact erasing non-vol without a backup to restore can result in a permanently crippled device which has lost some functions programmed at the factory and that effectively turns a senior fully optioned device into a junior less-optioned model. I know little more about Garmin's non-vol memory other than it's encrypted with simple XOR but even when decrypted I don't know how to permanently re-write data to it - maybe someone outside Garmin does but if so they've kept quiet. Seeing you've cleared NV previously i can comfortably say that either the non-clearable data in rgn154 is poisoned or Garmin's stuffed up with UK/ROI and Southern Europe subregions for these devices. However, seeing GN has been able to replicate your problem my money's on Garmin have stuffed up.

The reason the older UNL works with a newer version of the UK/ROI map is the same as why the properly generated code for the full EU supplied by GN works. The file contains the correct UID of course but crucially the correct FID of 12956 instead of "1" as in the GE supplied file and isn't dependent on the map version as such. You can try yourself, use JM to generate another random code for your UID with FID 12956 and i bet it'll work for your 2026.20 UK/ROI image file. The link is in Post #16.