…tions

Improved cohesion: worker catches browser exceptions and indicates
the kind of error to the communicator, which is the sole responsible
to deliver the appropriate error message to the end user and restore
the screen/cursor.

Worker indicates the error via a code (int) for now, but we might
consider using a dedicated ErrorReport object that includes more
information, depending on the type of error (eg rccode, inst, IP
address, html dump, stacktrace, ...)

Also kicked out much of the logout functionality for readability
purposes. We'll re-introduced logout to the code base after the
beta milestone.

This refactors the single parameterized method into a set of _explicit_
methods for each kind of expected failure, with additional info passed
as parameters to the communicator. Likewise, _unexpected_ failure (i.e.
exceptions) are handled via a single 'eventFailureInternalError'
communicator method that takes the traceback as an argument.

This nicely decouples visualisation concerns from worker functionality,
and ensures all kinds of failures are handled gracefully.

Use the Python atexit module to ensure the eventExit() method of
the running communicator is always called. This nicely allows a
communicator to for example redraw the cursor (cf issue #9).

Also improved the readibility of the browser URL requests, and added
a timeout for all remote server requests.

The lazy importing mechanism only imports a communicator when it has been
requested by the front-end, via the fabriek. This ensures an end-user
does not have to install all the dependencies for all the communicators
in order to run kotnetcli. The front-end is thus responsible to catch
any ImportErrors thrown at communicator creation time.

This nicely maintains the behavior that *only* the front-end and the
communicators encapsulate user interaction + only the front-end or
worker may terminate (sys.exit) the program.

This fixes issue #9 .