If I have a source database that is exposed using an API, would the normal pattern be to have a stream for that system, and then child streams for each "table"? I've been looking at example projects, but haven't found this pattern yet.

The common pattern for APIs is one stream for each table/entity (i.e. bulk

GET

endpoint). If you can query each table without additional information “from the system”, then they don’t need to be child streams.