Respect connect_timeout

[ansd]

Prior to this commit, when the client created a TCP connection with the
server and the server did not respond with a CONNACK (for example
because the server is blocked due to overload), the client was stuck in
state waiting_for_connack without respecting the configured
connect_timeout.

[zmstone]

Hi @ansd
Thank you for the PR.
Could you please rebase and add a line to changelog.md ?

[ansd]

Thanks @zmstone.
I rebased and added a line to the changelog.

[zmstone]

The problem, seems to be the use of event_timeout()
here:

emqtt/src/emqtt.erl

Lines 779 to 780 in fdd9959

	{next_state, waiting_for_connack,
	add_call(new_call(connect, From), NewState), [Timeout]};

and here:

emqtt/src/emqtt.erl

Line 834 in fdd9959

{next_state, waiting_for_connack, NewState, [Timeout]};

i.e. any event will cause the timer to be cancelled.
adding another event_timeout() after handing the event may seem to fix the issue, but maybe not the optimal way.

Should maybe use a state_timeout() instead. i.e. the total time allowed to stay at the waiting_for_connack state.

[zmstone]

A side-note:
Arguably, the timeout should include both the time to establish the TCP connection, and the time to send CONNECT and receive CONNACK.
However, if we are to fix it, it becomes a behavior change which is slightly incompatible to the previous versions.

[ansd]

Very good point and thanks for the explanation @zmstone: a state_timeout() is much better here. I pushed one more commit.