Prejoin hook does not fail cleanly.

If you error within the PreJoin hook in microcluster and then try to rejoin you get this:
```
Membership consistency check failed after timeout: Microcluster node membership is inconsistent across core_cluster_members ([10.33.87.13:6686 10.33.87.53:6686]), truststore ([10.33.87.13:6686 10.33.87.199:6686 10.33.87.53:6686]) and dqlite ([10.33.87.13:6686 10.33.87.53:6686])
```

This is affecting some new https://github.com/canonical/microovn functionality. It is not urgent but a fix would be nice.

Reproduce the failure by returning an error from the PreJoin hook in microcluster and then attempting to rejoin. Trace the membership consistency check across core_cluster_members, truststore, and dqlite; done means the failed prejoin leaves membership consistent and the subsequent rejoin reports a clean error.

Written by the indexing model from the issue text.