Suggestion: pedantic string encoding management

In the same way as python3 removed automatic conversion from string to bytes, I would like to have a way to prevent ``publish/will_set`` from auto converting payloads.
My problem started with paho encoding python2 bytes to utf-8 event if it should not (I saw the PR to fix it), and I then tried to find if encoding was well managed in my application.

Now in my client I just sub-classed 'publish' to assert payload is not an unicode string, (and convert bytes to bytearray for the bug).
I tried taking care of encoding since the beginning but this made me see many places where auto-conversion allowed bad string handling in my code.

Also, in practice, paho is able to automatically encode to utf-8 but cannot, of course, decode automatically so the magic is not symmetric.

Ideas on how to implement it:

* Add a an attribute client to choose this mode or not.
* Add a an option to 'publish/last_will' that sets encoding to 'utf-8' by default (could be backward compatible) and add a ``decoded_payload(encoding='utf-8')`` method to ``iMQTTMessage``.
* Remove auto-conversion and break everyone's application.

I would even make the auto-conversion raise a warning when it is not respected.
Crashing would be problematic as it can happen dynamically on a really well hidden case.
But this is a maintainer choice with other problems in mind.