While reviewing #28331 we noticed a crash that wasn't caught by the test suite, but would very quickly happen on a real node. See #28331 (comment)

I was able to reproduce it by calling addnode twice and having the test suite wait a little bit. But this also involved hacking ThreadOpenAddedConnections to wait 1 second instead of a full minute. See Sjors@bc3995c

If the test suite could mock the delay in ThreadOpenAddedConnections we can add a regular test for this.

It would also be a good idea to test MAX_ADDNODE_CONNECTIONS, which should be a simple matter of spinning up ~10 test nodes.