compositor-fbdev: make re-enable less drastic

Destroying the whole output in reenable would cause list walk
corruption: the loop over output_list in session_notify() is not using
wl_list_for_each_safe so output removal would break it.

Creating a new output is also problematic as it needs the compositor to
configure it, but that probably saved us from another list walk failure:
adding the new output to be list while walking the list, possibly
causing it to be destroyed and re-created ad infinitum.

Instead of a complete destroy/create cycle, just do our internal
disable/enable cycle. That will re-open the fbdev, re-read the
parameters, re-create hw_surface, and reinitialize the renderer output.

A problem with this is if fbdev_set_screen_info() fails. We do read the
new parameters, but we don't communicate them to libweston core or old
clients.

However, it is hard to care: to trigger this path, one needs to
VT-switch to another fbdev app which changes the fbdev parameters. That
is quite difficult as VT-switching has been broken for a good while for
fbdev-backend, at least with logind. Also fbdev_set_screen_info() would
have to fail before one should be able to tell something is wrong.

The real reason behind this patch, though, is the migration to the
head-based output API. Destroying and re-creating an output really does
not fit that design. Destroying and re-creating a head would be better,
but again not testable in the current state.

Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>