[FlinkRunner] Core should not follow remote flink runner job to completion

[zhilingc]

Expected Behavior

Core should submit the jar to the flink cluster and then the subprocess should terminate, instead of following the job to completion.

Current Behavior

Running a job using the FlinkRunner is a blocking call that runs indefinitely (streaming) or until completion (batch)

Steps to reproduce

• Run a job on a flink cluster

Possible Solution

Should follow the same process as DataflowRunner: submit job -> return job id -> allow the jobmanager to keep track of status of the job

We do need to retrieve the Flink job id from the Flink cluster in that case.

[zhilingc]

Starting a job in detached mode is not currently supported by Beam for Flink.

Runs are made detached by setting setDetached on the Flink ClusterClient. The option to do so is not exposed using the beam API. Ideally this would be a flag we can set so that any job (batch or streaming) can be run in detached mode.

Related issue is here: https://issues.apache.org/jira/browse/BEAM-593