"Socket closed" Spurious GRPC errors in Flink/Spark runner log output

We often see "Socket closed" errors on job shutdown, even though the pipeline has finished successfully. They are misleading and especially annoying at scale.

ERROR:root:Failed to read inputs in the data plane.
...
grpc._channel._MultiThreadedRendezvous: <_MultiThreadedRendezvous of RPC that terminated with:
status = StatusCode.UNAVAILABLE
details = "Socket closed"
debug_error_string = "{"created":"@1578597616.309419460","description":"Error received from peer ipv6:[::1]:37211","file":"src/core/lib/surface/call.cc","file_line":1056,"grpc_message":"Socket closed","grpc_status":14}"

Imported from Jira [BEAM-9082](https://issues.apache.org/jira/browse/BEAM-9082). Original Jira may contain additional context.
Reported by: ibzib.
Subtask of issue #20148