Better solution would be to make sure interface threads lock their object ALL THE TIME, except when waiting on its condition variable, so we do never "loose" condition signals. However, I am not sure if this would not result in a deadlock, should the interface need to lock another object (typically, playlist, input and/or vout).  Also, we would need to have all thread call vlc_thread_ready(), otherwise, we could also miss the signal if it happens right after the thread is created => this requires fixing every interface, except those that don't have a pf_run.

dbus: monitors input "state" variable to emit the TrackChange signal only when the metadata has been read

- Enable building of libamrnb and libamrwb (disabled by default). To enable add HAVE_AMR to config.mak and make sure to rebuild ffmpeg too.
- Bumped up libdvbpsi version number to 0.1.6

configuration: add config_AddIntf() config_RemoveIntf() and config_ExistIntf() to handle extra interfaces in the preferences
make use of it for last.fm simple preferences (ref #1262)
by the way the duplication of "control" with "extraintf" is very boring

qt4: delete the QApplication when we are finished using it, removes an ugly QObject::killTimers warning.

playlist_Clear() should delete only the local playlist, not the root playlist (including the media library, and the read only services discovery)

remove warning about libvlc and libvlc-control not being valid plugins

extras/contrib/src/Makefile: Fix dca an zvbi install name (on Mac OS X in particular). (Reported by maru).

We now have:
 1x 1.5x 2x 3x 4x 8x (and the same for slower).

MacOSX/Framework/VLCLibrary.m: Regroup the sharedLibrary allocation and release. Fix a warning due to a left over in previous commit.