Geo: update documentation to recommend adding secondary nodes to the database before starting them

Currently, the documentation asks users to set up a node and then create a record in the database (geo_nodes) that will cause the node to start acting as a secondary, based on its external_url.

However, this leads to error messages from the sidekiq and unicorn processes on the secondary as they try to act as a primary and write to the read-only replica database when they shouldn't.

By configuring the secondary in the database before installing them and starting those processes, we can avoid the situation and so make real errors easier to debug.