restore: full cluster restore leaves password authentication in inconsistent state

**Describe the problem**

After a full cluster restore, authenticating as a SQL user is inconsistently handled between different nodes in the cluster.

**To Reproduce**

Start with two multi-node clusters, each with a SQL user with the same name and different password.

Take a full cluster backup on the first cluster, and restore it into the second cluster.

Attempt to authenticate as the SQL user into each node of the second cluster. Some nodes will only accept the password from the first cluster (the "post-restore" state). Others will only accept the password from the second cluster (the "pre-restore" state).

**Expected behavior**

All nodes in the second cluster should only accept the password from the first cluster post-restore.

**Additional data / screenshots**

Restarting the nodes which accept the pre-restore password fixes the issue.

**Environment:**
- CockroachDB version: `v22.2.7` and `master`

Jira issue: CRDB-26474