Rethink bookkeeping of PipelineStage per pipeline

[jenny011]

In #317 , a PipelineExecutor instance keeps track of a PipelineStage which is reported to the client. The PipelineStage is updated and used not only in PipelineExecutor, but also in the GRPCHandler (instantiated per PipelineExecutor) and relevant code that are part of the pipeline logic.
Currently, the GRPCHandler and relevant code has no access to the PipelineStage stored at the PipelineExecutor. However, the GRPCHandler and relevant code also reports a pipeline stage to the client. This pipeline stage is hardcoded.
We would like to refactor the code so that

1. the PipelineStage, if reported to the client, is always bookkept at the PipelineExecutor instead of hardcoded.
2. logic about the PipelineStage is not scattered in multiple places.

[robinholzi]

Partially addressed: We now maintain the PipelineStage centrally through PipelineExecutor decorators. However, ambiguity still exists in the GRPC client interface. I'm not sure how much value it provides to also fix this.