Trigger failed_auth as before, once for each failed method
Trigger failed_all_auth once all method failed
Trigger no_auth only if we couldn’t even try one method

This is a big-and-dirty commit with a bunch of cleanup, maybe breaking a few
things, and not fixing all iq.send() calls yet.

Remove some useless things (like handling signals, managing the threads,
etc), add some comment to recently added/fixed methods…

All the auto_reconnect, connect_retry logic and that kind of stuf has been
entirely removed.

Scheduled events, connection, TLS handshake (with STARTTLS), read and write
on the socket are all done using only asyncio.

A lot of threads, and thread-related (and thus useless) things still remain.
This is only a first step.

This is behind a use_dnspython flag, however, so it can be disabled as
desired.

Fixes log.debug message in _connect_proxy

Fix support for jabberd2 with GSSAPI

Despite http://xmpp.org/rfcs/rfc3920.html#rfc.section.6.2, jabberd version 2.2.14 cannot accept the typical "<response xmlns="urn:ietf:params:xml:ns:xmpp-sasl">=</response>".  Instead it must be truly empty, so we force an empty response for this stanza only.

When the GSSAPI mechanism's process() function is invoked for the third time (on success) it must not attempt further processing.  Instead it should clean the context and return an empty response.

Replaced the reference to kerberos.GSSError.message in any raised exception, because:

 DeprecationWarning: BaseException.message has been deprecated as of Python 2.6

and its natural repr is probably the most desirable output.

Fix Don't process vCard avatars for MUC occupants caused TypeError