fix monitor disable test failures

[claws]

This change simply wraps the disable_monitor coroutine in a asyncio.Task to ensure it is run.

This change fixes the test failures observed. With the recent change to convert disable_monitor to a coroutine some of the monitor related tests no longer pass. However, this is not observed on Travis CI because it uses a libzmq and pyzmq combination that do not support the socket monitor capability - which in turn results in aiozmq monitor tests being skipped.

One strategy that might be worth considering in this area would be to use asyncio.gather to accumulate intermediate futures and then call connection_lost once the gather future fires. This would avoid the potential for race conditions of calling self._loop.call_soon(self.call_connection_lost, None) before any prior async functions complete.

[asvetlov]

Good catch.

I've converted disable_monitor to coroutine because I've found much easier to remember that enable_monitor and disable_monitor have the same signature.

I don't like running task for monitor disabling in close method. It's error-prone due undefined execution order. Let's add private non-coroutine _disable_monitor method and call it from both disable_monitor and close.

[asvetlov]

Oohh. I've missed your last changes, sorry.
Published as 0.7.1 release