release v0.23.8

Deactivate the stream in Cancel().  This fixes stuttering after a
manual song change by refilling the whole ring buffer before
reactivating the stream.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/1354

Clear not only MPD's ring buffer, but also libpipewire's buffers, to
avoid playing some audio from the previous song after a manual song
change.

Fixes part 1 of https://github.com/MusicPlayerDaemon/MPD/issues/1354

Removing the LockHasClients(); this code was copied from the "httpd"
output plugin, but unlike "httpd", the SnapCast output plugin does not
feed silence while paused, so we need to implement a delay to avoid
busy-looping the CPU.

As a side effect, this eliminates the suttering after resuming
playback, because the timer now gets reset even if there is a client.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/1394

Closes https://github.com/MusicPlayerDaemon/MPD/issues/1484

After connecting, PipeWire sometimes sends SPA_PROP_channelVolumes
with no values, and this led to "volume=-NaN".

Since PipeWire 0.3.53, there is no control name anymore, therefore the
name check doesn't work anymore, breaking volume change events.

This obsoletes the crash bug fix in commit 2ee57f9b

Fixes a potential crash bug.

This is a lower-level function without some of the clutter of
PipeWireOutput::SetVolume() which is not needed in that case.

This disables a long delay for playing songs from the cdio_paranoia
input plugin if ReplayGain is enabled.

The Read() method is not required to fill the whole buffer.  By
returning as soon as at least one byte was read, we allow faster
cancellation.

If seeking fails, don't leave the class with a wrong offset.

Use "size" instead.

cdio_get_devices_with_cap() can return nullptr if no drive was found,
or it can instead return an empty list.  The latter caused MPD to
crash.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/1532

If the PipeWire output has not yet been enabled and no thread_loop has
been created yet, a nullptr dereference in SetVolume() was possible
because nullptr was passed to pw_thread_loop_lock().

If a mixer is not open, rethrow the original exception each time
setting the volume is requested.  This further improves error messages
sent to MPD clients.

If a mixer is "global", it is available even if the output isn't
open.  However, since the check was changed from IsEnabled() to
IsReallyEnabled(), enabled outputs have not yet been used have not
been "really" enabled yet, preventing using the mixer.

Fixes a regression by commit 35dbc1a9
(part of https://github.com/MusicPlayerDaemon/MPD/pull/1480).

Closes https://github.com/MusicPlayerDaemon/MPD/issues/1563

This reveals more about the nature of an error instead of just
returning "problems setting volume".

libfmt version 9 broke the API by removing fmt::make_args_checked().

Fixes https://bugs.debian.org/1014543