0.4.5 spends 10 minutes at 100% cpu when loading old cached-descriptors* files

I started up an old relay that had cached-descriptors* files from a couple years ago.

It maxed out its cpu and sat there for around 10 minutes on a fast cpu, doing whatever it was doing. It ignored kill attempts other than kill -9. Eventually it finished and things were fine after that.

I tested and both Tor clients and relays encounter the issue. Tor 0.4.4 does not encounter it. Maybe it is related to the #40281 (closed) / #40221 (closed) changes?

It's not the end of the world, since it only needs to happen once per upgrade, and maybe if you have recent enough files it doesn't happen to you. But I would feel pretty sorry for people who don't have great CPUs and who get bit by this one.

I'll put the cached-* files up somewhere once I have a ticket number.