We should integrate with monitoring events from mongocxx to help debug failing atlas connections. Here is one example API we can look at, though we should look around the documentation/code for other sources of debug information.

https://mongocxx.org/api/mongocxx-3.9.0/classmongocxx_1_1v__noabi_1_1events_1_1server__closed__event.html

While we're doing this, see if there's any significant to the message "`serverSelectionTryOnce` set". Do we need to set any additional retry logic in the driver to try selection a server more than once?