objects: Make sure that we don't loose a vlc_object_kill() when attaching to a dead or dying parent.

Note: We do need to lock the parent for that.

since the sensor doesn't want to work on my laptop)

CID 140

Reported by Coverity scan : CID 142

Better patch to fix bug introduced by 0448c670 See: http://dinauz.org/~jpeg/VideoLAN/capture32.png (after left-click in the video output)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

I should do a real fix within a short time
Probably using vlc_gc_* for picture_t

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

Also fix the threads entry point prototype on Windows.

This reverts commit b8d7dae4.

There was a reason for the removal of this hack. I introduced that hack while trying to fix that precise compile error, and decided that this fix was hiding a bigger problem. Would be nice to summon before reverting, that could prevent such mistake.

While reading this, you will find a bunch of:
  while (find (VLC_OBJECT_INTERFACE))
     release; release;

These are of course plain BUGS (which are neither introduced nor fixed
by this commit). Imagine, for instance, what happens if two threads run
the code above at the same time... they end up releasing the interface
once too many.

Speeds up object creation and deletion, slows down vlc_object_get (which
you should not use anyway, remember), makes no difference for the rest

Revert "Remove wrong check for img_resample() when ffmpeg is built with libswscale. img_resample() is on the list of deprecated functions and is currently mutual exclusive with swscale." It was the wrong fix.

This reverts commit e1629c21.