Long workflows result in extremely high data transfer load

[mchl-hess]

The entire workflow graph is updated and stored with every chained/grouped task in the queue. This scales exponentially which causes significant slowdowns for very large workflows.

[pencil]

This has been addressed by several optimizations and a note in the README in 0.2.0.

A future optimization could be to store the metadata of messages in a separate store (e.g. DynamoDB) to further reduce the size of the queue.