Is your improvement request related to a feature? Please describe (👍 if you like this request)

Currently, the user can never finish draining a node if it contains a stopped replica of a volume and that replica is the only replica of the volume.

In some cases, that behavior is desired. For example, the user is trying to drain the node before replacing it. In this case, we block the drain to signal the user that it is not OK to remove that node since there will be data loss.

However, in some other cases, it is not desired. For example, the user is trying to drain a node and do an in-place node upgrade. In this case, we should not block the user's draining process because:

1. There is no harm to data integrity
2. There would be no data loss because the node is not going to be removed

Describe the solution you'd like

Create a new setting so that Longhorn removes PDB for instance-manager-r that doesn't have any running instance inside it

Describe alternatives you've considered

Manually evict the replica from the node before draining. This one is costly and not possible in a single-node cluster