Flink provides a migration guide:
https://nightlies.apache.org/flink/flink-docs-release-1.20/docs/dev/datastream/dataset_migration/
It is not trivial to port every operator, but the standard operations like map, filter, reduce and join should be easy to port.
Providing this as a seperate platform called flink-stream might be a way to ease development as you could avoid rewriting the FlinkExecutor and "just" write a new one from scratch that is built with DataStream in mind.