Reduce latency by stopping all AudioOutputs asynchronously.

Reduce shutdown latency by two commands per output.

Use ScopeLock to manage the lock; don't unlock after obtaining the
"really_enabled" flag; keep the same lock during EnableWait() /
DisableWait().

Without this, the pipe would run empty very often, which may result in
an xrun if the roundtrip to the PlayerThread and back takes too long.
By waking up the PlayerThread before the pipe runs empty, we make MPD
much more latency tolerant, which is a major optimization.

If we have a pending command, then the player thread is already awake.
Even if not, we'll wake it up as soon as we have finished the command.

Simplify the code, eliminate duplicate rollback code and eliminate the
"goto" kludge.

Now that I can see the Reopen() method clearly, it is obvious that the
out_audio_format setting will get overwritten by the following Open()
call.

release v0.19.21

The user unit omits the "ProtectKernelModules" setting which fails
with modular kernels:

 Failed at step CAPABILITIES spawning /usr/bin/mpd: Operation not permitted

It is unfortunate that systemd (version 232) is unable to reduce its
own capabilities, because this requires us to split system and user
units.

 https://bugs.musicpd.org/view.php?id=4608

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."

Yet another global variable removed.