A data engineering team needs to build a serverless pipeline that handles both batch and streaming workloads with a single unified programming model, and they want the resulting code to remain portable across clouds, on-premises, and edge devices. Which characteristic of Dataflow directly supports this portability requirement?