ep-engine ejects most active items to disk because memory is consumed by all dead vbuckets which janitor is not cleaning up because of rebalance failures

in this case we saw that there are 170 dead vbuckets in the memory and some of these vbuckets have 14k items in their open checkpoint which seems like a very large number.
the ep-overhead increased dramatically during rebalancing due to this issue as well but didn't go down because rebalancing failed in the cluster for som eother reasons.

active resident ratio was 0 percent at the end of the test run so user could not attempt another rebalance to fix this.

vb_1000:last_closed_checkpoint_: 17816
vb_1000:num_checkpoint_items: 14889
vb_1000:num_checkpoints: 2
vb_1000:num_items_for_persisten: 13574
vb_1000:num_tap_cursors: 0
vb_1000:open_checkpoint_id: 17817
vb_1000:persisted_checkpoint_id: 17814
vb_1000:state: dead