Currently the analysis step marks log lines such as "transition to primary complete" as an error (see here). However, for the upgrade/downgrade workloads where we will restart a cluster, that log line is expected and should not result in an evergreen error. As a result, we need to be able to override the `bad_messages` list in the analysis step to remove expected errors, or make them acceptable for our tests. 

See an example patch where this is failing.