Juju doesn't clean up all resources on remove-application

`juju remove-application` doesn't fully remove charm resources. For example, after deploying and removing metacontroller-operator, its pod and statefulset remain in place.

Environment:
microk8s 1.21/stable
Enabled addons: dns storage
juju 2.9.31

Steps to reproduce:
1. Add a new model named `test`
2. Deploy metacontroller-operator. Note: currently, a NotImplemented error is purposely raised on remove event, but in this scenario it has been removed and the charm was built locally.

3. Check the k8s resources:
```
$ kubectl get all -n test
NAME READY STATUS RESTARTS AGE
pod/modeloperator-6dfc6947bc-9btvg 1/1 Running 0 107s
pod/metacontroller-operator-charm-0 1/1 Running 0 44s
pod/metacontroller-operator-0 1/1 Running 0 67s

NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
service/modeloperator ClusterIP 10.152.183.189 <none> 17071/TCP 107s
service/metacontroller-operator ClusterIP 10.152.183.105 <none> 65535/TCP 69s
service/metacontroller-operator-endpoints ClusterIP None <none> <none> 69s

NAME READY UP-TO-DATE AVAILABLE AGE
deployment.apps/modeloperator 1/1 1 1 107s

NAME DESIRED CURRENT READY AGE
replicaset.apps/modeloperator-6dfc6947bc 1 1 1 107s

NAME READY AGE
statefulset.apps/metacontroller-operator-charm 1/1 44s
statefulset.apps/metacontroller-operator 1/1 67s

```
4. Remove the charm and verify if its resources were cleaned up:
```
 kubectl get all -n test
NAME READY STATUS RESTARTS AGE
pod/modeloperator-6dfc6947bc-9btvg 1/1 Running 0 5m26s
pod/metacontroller-operator-charm-0 1/1 Running 0 4m23s

NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
service/modeloperator ClusterIP 10.152.183.189 <none> 17071/TCP 5m26s

NAME READY UP-TO-DATE AVAILABLE AGE
deployment.apps/modeloperator 1/1 1 1 5m26s

NAME DESIRED CURRENT READY AGE
replicaset.apps/modeloperator-6dfc6947bc 1 1 1 5m26s

NAME READY AGE
statefulset.apps/metacontroller-operator-charm 1/1 4m23s

```
There are no errors in debug-log.