fix(rollingops): dynamic cluster id integration with etcd and sync lock

[patriciareinoso]

Description

This PR fixes a couple of issues found in the rollingops library:

1. Dynamic cluster ID integration with etcd does not work well.

Turns out that the certificates needed to connect to etcd were not created in case the cluster ID is not provided since deployment. Now they are created when the _EtcdRollingOpsBackend is instantiated.

Fix #480

2. The sync lock had a weird behavior in case of exception raised during the execution of the critical path. Now:

• we separate the exception handling between lock acquisition and execution of critical path
• the certificates to communicate to etcd are not removed during relation broken with etcd, and we do not fall back to peer relation back end on relation broken. This allows the unit to still use etcd during the teardown phase.

If the relation was removed from the application and we are not in a scale down situation, the rollingops manager will automatically fallback to peer backend later because the relation would be missing