Instead of creating a new eventfd for each DeferredMonitor instance,
reuse EventLoop's eventfd, and add a std::list to EventLoop that
manages the list of pending DeferredMonitors.  This std::list is
protected by the same mutex as the "calls" list.

The bottom line is: reduced overhead because the per-instance eventfd
was eliminated, slightly added overhead due to Mutex usage (but
negligible), and we're thread-safe now.

This subsystem is now good enough to replace EventLoop::AddCall().

Add the TagStream.cxx library, similar to TagFile.cxx, and use it to
load tags from song files inside archives.

Merge some duplicate code.

Got lost in commit c97685fe

Replaces decoder_plugin_from_suffix().

.. instead of decoder_plugin_from_suffix().  This reduces overhead by
walking the array only once.

Instead of using the first DecoderPlugin that supports the suffix, use
the first one that actually implements the "container_scan" method.

.. instead of decoder_plugin_from_suffix().  This reduces overhead by
walking the array only once.

If we have libyajl 2.0.1 (without a pkg-config file), our configure.ac
would assume this is the libyajl 1.0 API, because the function
yajl_alloc() exists in both.  This commit changes the library check to
the function yajl_parse_complete() which was removed in the 2.0 API.
This fixes build failure with libyajl 2.0.1.

We can assume that initgroups() would be a no-op in that case, however
initgroups() is not allowed for unprivileged users anyway.

Use the existing PcmDither library.

Instead of just adding a rectangular random value before shifting back
to the normal scale, use the existing PcmDither library.

This code is unable to dither.  Until we implement that, let's remove
the code for now.  i386 isn't relevant anymore anyway.

16 bit is not enough for volume calculations.