Ok, this is not a regression but a problem in the test script or rather
in the tests/network/sshlib's function init_remote_repo: it creates a
darcs repo remotely via ssh which means that our test-local
.darcs/config is not in effect. This causes the remote repo to be a
darcs-2 repo even if we test for darcs-1, since the mechanism we select
darcs1/darcs2 is via the defaults file.