When launched from systemd, kea needs to be restarted twice after a DB crash

In support ticket #15937, a customer reports:

During my testing of failure scenarios I found that when the DB backend goes away, it causes the KEA service to fail (expected and desired behavior). But once the DB is back online, the very first restart of KEA service fails with "address already in use" messages.

Additional supporting evidence is available on the linked support ticket but I have omitted it here because it has not been sanitized to preserve the privacy of the customer reporting it.