galera: avoid replacing resources

[dirkmueller]

Any restart of the database is going to cause a large downtime, so favoring availability we tell pacemaker to do this as little as possible. Also expire failures after a time of 30 minutes so that the clustering can resume after transient failures.