We should move this. It is a prerequisite to moving camerabin.

Some comments/questions after reviewing:

- GST_PHOTOGRAPHY_AUTOFOCUS_DONE: Name of custom GstMessage that will be
  posted to #GstBus when autofocusing is complete. Shouldn't we have a
  gst_photography_is_autofocus_done_message() for this, just like we have
  gst_is_video_overlay_prepare_window_handle_message() and
  gst_is_missing_plugin_message(), plus _parse_autofocus_done_message()?

- GST_PHOTOGRAPHY_SHAKE_RISK: same as for AUTOFOCUS done

- GST_PHOTOGRAPHY_PROP_*(): do we really want to require interface implementors
  to implement all of these properties? And then again implement setters as
  vfuncs? Seems a bit tedious, and also makes it more difficult for implementors
  to only support some of the functionality, doesn't it? (properties are good
  for things like automatic UI dialogs of course)

- struct GstPhotographySettings: used to set/get all options in one go. Assumes
  implementor supports all options; duplicates other mechanisms of getting and
  setting options (wrapper API for interface vfuncs and properties) with a third
  option, and the wrapper func doesn't even have fallback code for when the
  interface doesn't implement ::set_config or ::get_config.

- struct GstPhotographySettings: if we keep it it needs padding

- prepare_for_capture: why is the callback not done via a signal instead?
  (also, no destroy_notify for user_data, but callback can do that I suppose)
  How is failure communicated? Should we differentiate between still image and
  video capture?

- typedef void (*GstPhotographyCapturePrepared) (gpointer data, const GstCaps *configured_caps);
  why not GstPhotography-implementing object as first argument?

- set_autofocus: no gboolean return like all the others?