Retrieve documents from this knowledge store.
First, k nodes are retrieved using a vector search for the query string.
Then, additional nodes are discovered up to the given depth from those
starting nodes.
Optional metadata to filter the results.
Additional keyword arguments.