Windows CI does not fail if one or more tests fail

[stephengtuggy]

While I was working on updating the Windows CI, I noticed that some tests were failing, but the test suite -- and the GitHub Actions workflow as a whole -- succeeded anyway. We should probably change that.

[stephengtuggy]

OK, maybe I was mistaken. All the errors and failures I'm seeing now are expected errors or XFAILs. So everything is passing after all, it seems.

[stefanseefeld]

faber prints a test summary with overall stats that should indicate how many real errors were observed in the run. (And it is supposed to fail the overall call if unexpected failures were encountered.)