Improve health signal for StreamingDataflowWorker

The StreamingDataflowWorker currently just uses the health of the FnAPI server for its own health monitoring. It should also take into account persistent failures to process work items successfully which may be due to a bad worker. For example, a recent Snappy initialization error was due to /tmp becoming full.