SafeNotifyService threads leak in CuratorFrameWorkImpl

[fapifta]

CURATOR-495 introduced a new runSafeService field in CuratorFrameworkImpl class, and this field is either initialized by an external ExecutorService via the builder, or it is created internally within the class.

In the CuratorFrameworkImpl#close method though, this Executor is never closed, so the threads that are opened by the instances are lingering there until the VM is closed by default. Worse, if someone specifies a thread factory to the framework implementation via the builder that produces non-daemon threads, the VM never exits due to the unstopped single thread executor.

[fapifta]

I have posted a PR(#1278) that adds tests to demonstrate what I think is the expected behaviour, please let me know what you think.