Again, I'm not quite sure it's supposed to work, but it seems reasonable: if the server can discover a table on appearance of ARZ file and rediscover it on modification of ARZ file, then why not to anti-discover it when the ARZ file is gone? If it's by design, please close.
The test below is only one line different from a fragment in archive.discover test.
In archive.discover, the select under ER_NO_SUCH_TABLE is done from t1 (which truly doesn't exist), not from t0. I don't know if it was intentional or not.
This test fails with
ER_NO_SUCH_TABLE would have been much more elegant.