This eliminates some overhead, because the compiler doesn't need to
consider these functions throwing.

https://bugs.musicpd.org/view.php?id=3876

From the avformat_open_input() API documentation:

 "Note that a user-supplied AVFormatContext will be freed on failure."

https://bugs.musicpd.org/view.php?id=4607

Replaces decoder_data() and others.

Replaces decoder_initialized().

Prepare for a Decoder API redesign based on an abstract class with
virtual methods.

An empty packet would be a command for avcodec_send_packet() to
finalize the codec.

Fixes https://bugs.musicpd.org/view.php?id=4588

This function exists since FFmpeg 3.1.  Fix a build failure with
FFmpeg 3.0.

These bug reports describe problems with some FFmpeg codecs:

 https://bugs.musicpd.org/view.php?id=4564
 https://bugs.musicpd.org/view.php?id=4568
 https://bugs.musicpd.org/view.php?id=4572

According to the FFmpeg bug tracker, a call to
avcodec_parameters_to_context() is required after
avcodec_alloc_context3():

 https://trac.ffmpeg.org/ticket/5781

This requirement was previously undocumented.

Turns out the libavcodec version numbers are not linear;  the feature
was added in FFmpeg 3.1 commit 998e1b8, libavcodec 57.14.0; but FFmpeg
3.0 has version 57.48.101.  Ouch!

Was accidently added by commit cafc266e

This commit suppresses the remaining deprecation warnings with FFmpeg 3.1.

The AVCodecContext attribute is deprecated.

Preparing for FFmpeg 3.1 support.

Improve initial seek by not reading/decoding the first frame before
checking for the seek command.

Use the duration of the stream we're actually decoding - not the
"global" attribute AVFormatContext::duration which may differ.