When shutting down a ShardingTest, we often run validation hooks, like validating sharding metadata is correct. These hooks generally assume all servers in every started replica set are available, which isn't true in all tests. We usually work around this by manually marking tests to skip certain problematic validations, e.g. here where we manually disable 4 validations. Instead we can add an assertion that every member in ShardingTest is running when ShardingTest#stop() is called, forcing existing tests and any new ones to restart any nodes they shut down before ending the test, which may make tests run slightly longer, but should lead to fewer BFs and make adding new validations easier.