Use separate branches for master and data nodes

[jacobwinch]

At the moment Step Function executions for master and data nodes follow the same series of tasks:

If the node is a dedicated master-eligible node then certain steps (e.g. MigrateShards) are unnecessary.

It would be easier to reason about this project if there was a separate branch for dedicated master-eligible nodes, which skipped the superfluous tasks and checks which are specific to data nodes.