We require use of test_must_fail to check expected non-zero exit by
Git itself, but discourage test_must_fail to be used for checking
exit status of non Git commands that are supplied by the system.
The current text explains the reason for the former but not the
latter.