macos and windows tests are able to rot

Currently, our OSX build test are broken (#1394 (closed)). This has occurred because of a combination of the following factors:

1. These build tests only run on main, so fixing them is not treated as a priority because no-one's work is blocked (except for the release technician's)
2. Probably, these build tests are not hermetic. I think this because the errors in #1394 (closed) don't look like the kind of thing that would happen due to us updating arti.git. IMO changes made outside arti.git should not cause our own tests to fail.
3. Perhaps, those of us who have enough macos expertise to fix this don't have enough bandwidth.

As a result, we are not meeting the goals set out in our Support Policy. @ahf/@nickm, is there a label that reflects the seriousness?

(I've made !2226 (merged) which turns the build failure from a release blocker into merely a serious bug.)

(1) and (2) are what this ticket is about. If after that, we find that we can't keep macos builds working, (3) would be a matter for management.