Cancel scheduled ControlConnection refresh if shutdown was called

[betalb]

Store timeout, used to schedule refresh call, and clear it on shutdown.

Even though refresh call has check if shutdown was called, timeout itself may prevent nodejs process from shutting down until timeout is reached.

I have some concerns about test itself. Not sure that all the setup steps are required. But I've copied it mostly from previous test "should reconnect when all hosts go down and back up". What I wanted from the test:

• spin up cluster
• open control connection and ensure that it is live
• stop cluster
• call control connection shutdown
• verify that refresh was not called

So the main test is pasted below, the rest is a boilerplate copied from other test.

// start spying on _refresh calls
cc._refresh = sinon.spy(cc._refresh);

cc.shutdown();

// wait for reconnectionDelay with 10% lag
await helper.delayAsync(parseInt(reconnectionDelay * 1.1, 10));

assert.ok(cc._refresh.notCalled);

[SiyaoIsHiding]

Is this the only refresh timeout that needs to be canceled?
We probably should store all the refresh timeouts as controlconnection fields, and clear all of them at shutdown.

[betalb]

Need to refresh my memory. There is one more suspicious timeout created in _waitForReconnection, but I remember that I've spotted it as well. But, most likely, I've ignored it because it is small (2s).

I have a code that handles this timeout as well, but need to write a test for this particular case.

[betalb]

Ah, no, there was another reason why I decided not to handle timeout in _waitForReconnection explicitly. During shutdown newConnection event is emitted (with error) which triggers cleanup of this timeout. Probably small comment will make sense.