Readd tests for cache trimming priority in the [copy] mode

[Blaisorblade]

Those tests were removed in #4511, and #4497 (comment) already hinted at concrete problems.

[Blaisorblade]

FTR, using a cache entry by opening it read-only (as done in copy mode) should not update its ctime, unlike creating hardlinks — that's according to the POSIX standard (or Mac/Linux man pages).
https://pubs.opengroup.org/onlinepubs/9699919799/functions/open.html

However, I could still have missed code in restore_artifacts that touches permissions in other ways. I'll try to test this myself if nobody confirms the problem first.