system testing does not verify expected ADB behavior when doing overmem cleaning

As introduced on #2405, BIND currently does not behave well in at least some circumstances when the ADB is overmem and performing overmem cleaning.

Presumably we added the overmem cleaning logic with the intent that BIND should continue to run and operate relatively normally when the ADB goes overmem, and that such behavior would be better than other alternatives (e.g. flushing the entire ADB as soon as the overmem condition is detected).

We have not documented our expectations for BIND's operational performance when this behavior is triggered, nor do we have any system tests that attempt to verify that BIND is continuing to behave as we expect.