Currently there is no guarantee that the ShardingDDLCoordinator will emplace their completion promises before their destruction. In fact it is possible that we loose all the references to a coordinator, in fact if a stepdown/shutdown happens between:

1. By removing its stored document (the POS will remove its reference from the active instance map)
2. By interrupting the command from which the coordinator was originated.

We will loose all the reference to the coordinator that will be destructed without having the chance to set its completion promise.

The proposed solution here is to move the removal of the coordinator document to the onCompletion task of the ShardingDDLCoordinator, so that ti will be done in the same future chain task.