Upgrade to a version without the bug, or fix the bug yourself. (If I had this problem and was stuck on a 1.6 version (we actually are), I would look to see if it had been fixed in a later version and see if I could back port that fix. That is well beyond what you can expect from peer support.
The way to get rid of the caller is to hang them up. If the queue data structures are out of line with the channel ones, you have a bug, and you need to fix the bug.
In fact, if you do manage to force the queue entry to be abandoned (e.g. resetting its timeout - there is no user interface to do this), there is a good chance that you will get a restart now, as there is a good chance that the queue application will try to close the channel whose data structure no longer exists, eventually causing a memory corruption crash.
You can probably change the dialplan to queue new entries at a higher priority.