address concurrent issue with sleep future

[demoray]

As is, the Sleep future relies on the future not getting polled until the waker files, which isn't guaranteed.

This changes the implementation to check if the thread is done. The waker still triggers the polling to occur.

fixes #1170

[cataggar]

I'm not super familiar, but looks good to me. May be @rylev or @yoshuawuyts can review.

[johnbatty]

Looks good to me too.

You can see that in the original code that the sleep timer is started on the first poll, and if a second poll is made before the timer has completed it always returned Poll::Ready, which is incorrect. The fix correctly handles any polls after the timer has been started by checking for timer completion before returning Poll::Ready.