And got a fix for the issue. The timestamp wasn’t the problem, the auto-generated database code for that table was… When we added support for configurable image expiry, the DB filter was adjusted to filter on a per-project basis.
But the generated DB code wasn’t updated to generate the correct queries for a project+cached filter, causing the cached part to be ignored and it only filtering based on projects…
https://github.com/lxc/lxd/pull/8579 is the fix for this issue (and contains the full explanation of what happened).
Please note that anyone affected may still get re-affected by this until the bugfix is made available to all snap users! It will not hit you until 10 days since you last used a particular image, I intend to roll out the fix for this tomorrow to all stable users.
The fix has been merged upstream and cherry-picked into our snap packaging branch, I’ll now let it build and once built and CI passes on it, will push to stable, ETA is 2-3 hours.
Glad this got sorted in the end 
Took longer than expected because of some issues on the package builders requiring many many retries, but it’s now all good and in stable.
The database you sent me appears to be for a standalone system, not a cluster, is that correct or am I misreading the database?
Ah, yes my mistake this one was not clustered as you noticed. It was our newest and least utilized but still had the issue!
Wow I did not expect something like this to be fixed this fast! Proves again how excellent this project is. Thanks for all your help as always!