Refactored close procedures to dispatch Errors for all components

[tvoinarovskyi]

Basically now it will always raise appropriate exceptions for any public calls after closure. Added ClientClosed and ServerClosed exceptions. Maybe they are a little to grained, your thoughts?

Now we can write:

@asyncio.coroutine
def do_stuff(connection):
    queue = yield from declare_queue(connection)
    consumer = yield from queue.consume(lambda x: print(x), no_ack=True)
    yield from asyncio.sleep(1)
    queue.publish(some_message)

connection = yield from connect()
while True:
    try:
        yield from consume(connection)
    except asynqp.exceptions.AlreadyClosed:
        connection = yield from reconnect()

and it will raise no matter where connection is lost.

Depends on #54. Please review that firts. I kinda split them in a way )

[tvoinarovskyi]

Rebased

[tvoinarovskyi]

Bump. I would really appreciate feedback on this, as it's a stopper for #34

[benjamin-hodgson]

Apologies, I haven't had time to review this. I'll try and do it ASAP

[benjamin-hodgson]

Is this possible? If a user could provoke this condition by (ab)using the API, we should explicitly throw an exception. (If you believe it to be impossible, then an assertion is fine.)

[tvoinarovskyi]

I think it's not possible. We didn't specify protocol as part of public API in docs.

[tvoinarovskyi]

Squashed and cleaned code. Removed some unneeded changes, so they don't bother in review.
The idea to use PoisonPill for closing was quite helpful, I like this one more.

[tvoinarovskyi]

bump again