Thread leakage in Hadoop RollingAverages metric collection

The rolling averages metric implementation defines a static ScheduledExecutorService instance which is not shutdown properly. This results in leaking the corresponding thread. This issue was identified while testing Solr/Hadoop 3 integration.

Here is the relevant portion of the code involved,
https://github.com/apache/hadoop/blob/c1549352cf1d1b1b45d7b613f993d45649b8efcf/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/lib/RollingAverages.java#L65