GitLab

* we called unbind_for_thread_switch unconditionally, but this
  is wrong if the previous thread exited
* likewise, exiting a thread should clear current_thread
* redundant assignment in run_thread
* clean up init_threads - no need to re-init the primary thread

This patch still sometimes causes weird hangs in "make check".
However, I think that is a kernel bug, since Emacs enters the zombie
state but its parent process hangs in wait.  This shouldn't happen.