Run FontBuilder thread after Freetype has been initialized and no error paths are can be taken. This should solve the crash reported by several people with the video title (or meta-name) appearing at start of the movie.

The freetype module starts another thread called FontBuilder and depending on timing this creates a crash or works allright. The crash is triggered by the font not being found. This patch tries to load the font first before the FontBuilder thread is started.

I was unable to reproduce the crash with the patch applied. Without the patch the crash is reproducable.

As discussed on the mailinglist the FontBuilder thread doesn't depend on the freetype initialisation and rearanging the order should do no harm.

(there was no meta data support for ac3).

(We still parse the flac meta data as a safety, in case taglib support is not there)

if the demuxer has not set them.
 Added a demux2 control DEMUX_HAS_UNSUPPORTED_META.
 A demuxer should return true only when the format support extra meta
data (id3/ap2/...) but it does not read them itself. The input will then
call a "meta reader" (demuxer modifications on the way).

	  -get rid of bad merge remnant

contribs: don't try to install the shared library if it doesn't exist, fix [22437]. Spotted by John Dalgliesh

checks if playlist has been stopped in the input "state" callback
holds the p_sys mutex only when writing/reading audioscrobbler_song_t data

allocate as much as we can on the stack, to avoid numerous malloc() & free()
fix memory leaks by the same way
metadata reading is now performed exclusively from the callbacks
don't use linked lists anymore, but a queue of a fixed size of 50 songs, i.e. the maximum we can transmit in one submission.
properly unloads the module when encountering a fatal error, like no username/pw, no memory, VLC being banned by last.fm servers ...

homepage: use %? to represent '?'
other: probe is not successfull anymore for urls like http://10.0.0.1/get_video?xxxx&origin=youtube.com

      Cosmetics
      Set svn:keywords

Someone should really review this.
This does not affect the trivial resampler; I have not checked the others.