Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Rather than automatically expanding folders when hoving over them, use this
information (folder expanded or not) to either insert the dropped item as its
next sibling in the parent folder or as its first child in this folder respectively.

This patch includes the following :
  - fix item misplacement (items were only appended instead of being inserted)
  - fix slider scrolling that could not adapt to the real size of the playlist
  - enhance drag&drop by allowing users to finely insert item being dropped
    into either the playlist or the media library.
  - optimise refresh (only rebuild playtree in case of visible item)
  - remove keeping a reference to a playlist_item_t* (since it is not
    refcounted, a lookup from the playlist with proper lock mechanism
    is needed)
  - remove the m_deleted flag (corner cases were never dealt with) and
    replace it with a notification prior to deletion
  - implement operator++ to simplify iterating visible items (cosmetics)

A deeper redesign/simplification and support for the new sql playlist would be a good thing though.

Pass drap&drop coordinates over to visual controls.
This feature will be used to allow users to finely insert items in the
playtree control instead of just appending them at the end.

Hand-crafted code in plugin descriptor is not allowed. And the code was
unsurprisingly buggy.
This reverts commit 1c2a95b8.

Conflicts:

	modules/codec/schroedinger.c

macosx: fixed volume normalizer prefs and crash when saving the Growl option in combination with some config sets

It is a preliminary work.

It is looking more natural.

It allows to use hw decoding (whereas it was always disabled).
It workaround a bug in ffmpeg/libav.

It avoid tearing depending on the default settings.

 * Add a dispatcher method for tracklist properties
 * Implement Tracks and CanEditTracks

restricted to the audio and video filters.

This happen when p_vids->p_bih->biSize - sizeof( BITMAPINFOHEADER ) = 0
and the data is in p_vids->i_chunk_size, because then p_extra == NULL.

This solves the fruity sample crash but doesn't fix the palette