https://bugs.musicpd.org/view.php?id=4665

https://bugs.musicpd.org/view.php?id=4656 describes a crash due to
division by zero because frame.samples==0.  This should never happen,
but apparently can happen after seeking.  The best we can do is to
just ignore this frame.

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

The libWavPack documentation says:

 "After a FALSE return the file should not be accessed again (other
 than to close it); this is a fatal error."

Technically, these samples are signed.  There was no practical
difference, only the declaration was formally wrong.

Since we don't use OPEN_2CH_MAX, this should be equal to
WavpackGetNumChannels() - but who knows, maybe a future libWavPack
version breaks this assumption.

Enable OPEN_DSD_AS_PCM if available.  No OPEN_DSD_NATIVE yet.

 https://bugs.musicpd.org/view.php?id=4606

Use MPD's internal APE tag code, which works just as well.  This
removes some duplicate code for reading tags and ReplayGain.

MPD uses a floating point range from -1 to +1 internally - why ask
libwavpack to use -32768..+32768 only to reduce it back to -1..+1
afterwards?

It's wrong to use std::system_category() for both GetLastError() and
errno on Windows.  Apparently, everybody uses std::generic_category()
for errno values, which appears to be a safe choice.

Some discussion on this confusing topic can be found here:

 https://stackoverflow.com/questions/28746372/system-error-categories-and-standard-system-error-codes

Declare two named elements.  An enum should not be used as an array
index, as this is error prone.

Speed up container_scan() again, by eliminating the need to call
scan_file() for each item.