GitLab

* Killing the module before it had shown the hourglass used to remove the
  pointer.
* If an hourglass with a long lead-in time is pending, and a (shorter lead-in)
  second hourglass is started, the time that the animation actually starts is
  now determined by the sooner of the two, rather than always by the top-level
  hourglass' lead-in time.
* If something else changes the pointer number or colours while the Hourglass
  is animating, the new values are remembered and put into effect when the
  hourglass stops, overriding those that were active when the hourglass started.
* Under certain circumstances, R10 could be corrupted on exit, leading to
  data aborts being generated by the Wimp error box code. Fixing this also
  means that Hourglass_Smash can call Hourglass_Off directly, rather than
  having to issue Hourglass_Off as a SWI.