Added FlowControl to Protocol

[TarasLevelUp]

Now we can wait for writes if TCP buffers are full. This can be very useful for Publisher applications.
NOTE: not too sure if adding an option to publish is that good of an API. Any ideas?

[tvoinarovskyi]

Must probably be fixed based on #56

[JustinTArthur]

@TarasLevelUp I think exposing buffer flow control hints is great—thanks for getting the ball rolling. As for adding it as an option to publish, I don't quite like the idea.

One problem is using the word "flush" in the naming/docs. It gives the idea that the publish has been flushed to the system's networking stack and this isn't necessarily true. We might have flushed the buffer below the high watermark, but some of our publish might still be in the buffer unflushed.

Secondly, having two different return types for the same API method feels wrong.

What about exposing/documenting that drain coroutine you added to the sender object? If we don't document sender anywhere, we could have a coroutine on the connection object that calls it.

Something like

@asyncio.coroutine
def publish_numbers():
    connection = yield from asynqp.connect('localhost', 5672, username='guest', password='guest')
    channel = yield from connection.open_channel()
    exchange = yield from channel.declare_exchange('test.exchange', 'direct')

    for number in range(100):
    	msg = asynqp.Message({'myNumber': number})
    	
    	yield from connection.drain()
    	exchange.publish(msg, 'routing.key')

I don't like the term "drain" either, but it does match watch asyncio and nodejs call it :)