Scene-graph improvements

One of the things missing in scene-graph was having a reason (contextual human-form like) as to why the view could not been placed/assigned into a HW plane.

This series tries to address some the comment raised in #244, and, in some extent it should supersede !333, but only on the display part not the calculations further discussed there (specifically by e90d7bfb).

This series:

• adds a new callback into weston_output which can be installed by the backend than can be called from libweston: weston_compositor_print_scene_graph().
• we hang off a uint32_t reason into weston_view and use front-end weston_view_set_reason_for_compositing() in the backend set the reason
• the human-form strings are completely opaque to the front-end
• the backend-drm contains the string related reasons

Further it also adds a new scene-graph, scene-graph-by-output which only changes a bit how the views are displayed, by grouping the views first by the output, then in layers. It shows views for each output, but still keeps the layers numbering. Arguably it should be much more easy to get a sense of the scene-graph in multiple output set-ups. This might require some cleanup as it adds some duplicate parts, but not really sure if it makes to see how to do it better in case this is not something that seems good to have.