Also fix a use after free

POSIX sem_*() functions return -1 on error and the error code in errno
(contrary to most POSIX thread functions).

This gives a use to the otherwise useless artUrl variable
This prevents qt4 interface from consuming 100% of CPU when art image
has to be scaled from a big source picture

This reverts commit 4b5229c8.

The stream_out_t chain creation is modified: all modules are created by
the core (or by stream_out_duplicate) instead of being created by the
previous module.

sout_StreamChain{New,Delete} replace sout_Stream{New,Delete} to handle
modules chains instead of individual modules

sout_Stream{New,Delete} are still used by those new functions but made
static inside stream_output.c

Remove now unneeded psz_chain from struct sout_instance_t
Replace pointer to chain of next module by pointer to next module in
struct sout_stream_t

Example use:

vlc --sout-all input.mp4 --sout
"#duplicate{dst=transcode{vcodec=mp2v},select=es=0,dst=transcode,select=es=1}:std{...}"

(dst=transcode without acodec/vcodec is a hack to pass the encoded stream to
stream_out_standard without transcoding)

I vote that we remove this plugin now.
FFmpeg supports SIPR natively now.

Sorry

Win32: add support for play and pause buttons in media keys (different from a plya/pause unique button

Tested with an Apple remote v1, and a HP media center remote. I don't know how to get the "menu" message" from the remote.

Apparently libvlc_time_t is millisec, whereas mtime_t is microsecs.
Most event callbacks where carying an incorrect mtime_t value.

We could handle directories as descriptors, but we currently don't. In
the mean time, we had better show an error than generate corrupt
playlist entries.

"-" is now converted to "fd://0" by the input code. And file:///- really
means /- (which is a valid file name).

venc=x264{slices,slice-max-size,slice-max-mbs} options are there now

libx264 gives dts-value, but we need to calculate initial delay, so dts doesn't
go < VLC_TS_0.

Tell how many frames are still left on libx264 buffer when we close encoder

Only set BLOCK_FLAG_TYPE_I on IDR-frames on X264_BUILD < 83 and only those
which have b_keyframe on X264_BUILD >= 83.

We assume it is present in plenty of places anyway

This requires support for fdopendir(). One open() and fstat() calls per
input file are avoided. Ok, this is not such a major improvement).
This should also work around brain-damaged file system drivers such as
Linux HFS+, whereby opendir() succeeds on regular files.