Drivers could mark mongobetween Unknown for server issues

The README says that mongobetween appears as an always-available mongos server, but I believe a driver would mark mongobetween as `Unknown` if there were a server error like `NotMaster`. In this case, mongobetween would extract the error and use the Go driver's `ProcessError` function to mark the actual mongos Unknown, which is correct, but would also proxy the message back to the original driver, which would then mark mongobetween Unknown per the error handling section of the SDAM [spec](https://github.com/mongodb/specifications/blob/master/source/server-discovery-and-monitoring/server-discovery-and-monitoring.rst#not-master-and-node-is-recovering).

EDIT: I think the same is true for connection errors. `mongo.RoundTrip` returns an error if the `WriteWireMessage` or `ReadWireMessage` calls fail. This is propagated upward so `handleConnection` returns the error to the goroutine launched by `Proxy.accept`, which closes the connection. This would show up in the application as a non-timeout network error, which would cause the application to mark the proxy as Unknown and clear its connection pool.

Start with the README and trace the named flow through mongo.RoundTrip, WriteWireMessage, ReadWireMessage, handleConnection, Proxy.accept, and the Go driver's ProcessError. Verify how server and connection errors are reported to both mongos and the original driver; done means the expected SDAM behavior and any required change are clearly established.

Written by the indexing model from the issue text.