When a replset node is down, and the connection string provided on the command line includes that node, mongosh fails to connect.

This is a consequence of our fail-fast connect implementation from MONGOSH-705, where ECONNREFUSED|ENOTFOUND|ENETUNREACH are considered failures that should result in aborting the connection attempt.

Ideally, we should try to identify situations in which all servers we connect to give serverHeartbeatFailed failures, not any of them.

If that’s impossible, we should only do fail-fast for single-host deployments (although that would reduce the usefulness of MONGOSH-705 quite a bit).