The definition of AVOCADO_JOB_FAIL was "Something went wrong with the
Job itself, by explicit avocado.core.exceptions.JobError exception",
while AVOCADO_FAIL is defined as "Something else went wrong and
avocado failed (or crashed). Commonly used on command line validation
errors".

With those in mind, I support that AVOCADO_JOB_FAIL should be slightly
adjusted (see changes in the diff), to include other types of
failures, but limited to job *execution* failures.  AVOCADO_FAIL is
already generic enough, so there's no need to change its definition.
This is important, IMHO, because users could quickly understand if a
job was executed, and failed, or if some earlier type of failure (such
as command line validation) happened.

With this updated definition, I spotted a few places where AVOCADO_FAIL
and AVOCADO_JOB_FAIL usage should be changed.  Examples:

 * In `avocado.plugins.diff` the resultsdir is attempted to be read, but
   in my book that still doesn't qualify as a job failure, because a job
   is not being executed.

 * In `avocado.plugins.replay` while there's functionality to run a
   (replayed) job, many of the occurrences of AVOCADO_JOB_FAIL are
   similar to the `avocado.plugins.diff` and happen while trying to load
   jobdata from the previous job resultsdir.

 * In `avocado.plugins.multiplex` the situation is a bit more tricky.  The
   same code is executed when multiple avocado subcommands are executed.
   If the currently executed command is `run`, it would mean that a job is
   being attempted to be executed.  In fact, a job *is* already executing
   at this point, so a AVOCADO_JOB_FAIL is appropriate.  In other scenarios,
   such as when running `$ avocado multiplex`, these failures have nothing
   to do with jobs, so AVOCADO_FAIL is appropriate.
Signed-off-by: NCleber Rosa <crosa@redhat.com>