Bug #915 is about an I/O exception thrown where none was allowed,
leading to crash via std::terminate().  However, instead of catching
and logging the error inside the decoder plugin, it should be able to
propagate the I/O error to the MPD core, so MPD can avoid trying other
decoder plugins.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/915

The formula did not consider the channel count.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/639

These are used by DSD-streaming servers.  For example, MiniDLNA uses
"audio/x-dsd".

The scan methods must be "noexcept".

Replaces decoder_data() and others.

Replaces decoder_initialized().

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

Check for STOP before decoding the first chunk.  This reduces the
command latency.

This allows the compiler to discard buffer contents between two
iterations.

Nobody cares.

Prepare refactoring the whole plugin to use blocks instead of bytes.
A block is the smallest addressable unit, and it will simplify the
seeking code.

This finishes the multi-channel support.  Development of the feature
was started with commit 02cc77cd

When calculating the upper bound using the "sample count" format
header, don't assume it's stereo.

The plugin was horribly bugged for files that were not stereo.

Don't pass the buffer size to the function, as it's known at compile
time.  Use "restrict" on the pointer arguments, and merge the two
loops, which allows the compiler to optimize this loop with a few SSE2
instructions.

Read one block at a time.  This discards the last partial block, which
cannot be interleaved anyway.  Previously, uninitialised memory was
used to interleave the last block, which generated some noise.

Convert into a second buffer that gets passed to decoder_data()
without copying back to the first buffer.