In case the dependent Dataset is already computed/in storage, we want to update corresponding part of streaming graph from streaming nodes to source nodes (read from storage rather than re-compute)
This may require implementing a Features Metadata Store to see which values are computed/missing (similar to what Usman proposed)
This consists of 2 parts.
This may require implementing a Features Metadata Store to see which values are computed/missing (similar to what Usman proposed)