Java performance benchmarks dirty shutdown

https://grpc-testing.appspot.com/view/Performance/job/gRPC_performance_experiment/4231/consoleFull

In most of scenarios, the java qps_workers don't seem to shutdown cleanly:
```
2017-06-26 12:32:42,813 WAITING: ETA 66.3 sec; 0 queued, 1 jobs running, 1 complete, 0 failed
2017-06-26 12:32:42,860 E0626 12:32:42.832971879 11903 driver.cc:501] Worker 0 could not be properly quit because Received RST_STREAM with error code 8
E0626 12:32:42.847037641 11903 driver.cc:501] Worker 1 could not be properly quit because Received RST_STREAM with error code 8
E0626 12:32:42.859240520 11903 driver.cc:501] Worker 2 could not be properly quit because Received RST_STREAM with error code 8

2017-06-26 12:32:42,860 FAILED: qps_json_driver.quit [ret=1, pid=11902]
```

The log also has occasional `[GC (Allocation Failure)]` messages:
```
[GC (Allocation Failure) [PSYoungGen: 524800K->11962K(611840K)] 524800K->11978K(2010112K), 0.0112577 secs] [Times: user=0.04 sys=0.00, real=0.02 secs]
[GC (Allocation Failure) [PSYoungGen: 536762K->4660K(611840K)] 536778K->4684K(2010112K), 0.0062845 secs] [Times: user=0.03 sys=0.01, real=0.01 secs]
[GC (Allocation Failure) [PSYoungGen: 529460K->4580K(611840K)] 529484K->4612K(2010112K), 0.0056438 secs] [Times: user=0.02 sys=0.00, real=0.00 secs]
```

Start with the linked performance experiment console log and the qps_workers shutdown path associated with qps_json_driver.quit. Investigate the repeated RST_STREAM errors and whether the GC messages are related. Done means the Java workers shut down cleanly and the benchmark no longer reports a quit failure.

Escrito por el modelo de indexación a partir del texto del issue.