- Old (>4 hour) directories are deleted at the beginning of the
  execution, so the disk full problem should be out!
- There are 2 types of errors:
    - Controlled errors: The backup process finish although something
        has been wrong.
    - Uncontrolled errors: The backup process dies (memory fault, apache
        problem, electric problem...:-)).
- Now scheduled backup is able to detect previous uncontrolled errors to
  avoid repeating the same course always. It'll be deferred to the next
  execution.
- Every time that a backup cycle has finished, admins will receive an email
  with the summary of the execution.
- The 'ERROR' text will be present in email subjects if something was wrong.

With this, all should be ok (I hope).

Merged from MOODLE_14_STABLE