Eliminate the conversion to integer samples, the MPD core can do this
now.

For MPD's new floating point support: when a decoded wavpack file
needs to be resampled, don't convert float to int and back to float.

For future internal use.

Support for conversion from float to 16, 24 and 32 bit integer
samples.

Let the caller deal with a failure.

Don't suppress gcc warnings.

A version of the resampler that doesn't need int->float->int
conversion.

Reduce the mess with local variables inside switch{}.

Allow more compiler optimizations.

Reduces number of implicit integer conversions.

Didn't compile on GLib < 2.16.

This is easier and more efficient to loop on, because only two
variables get modified (src and dest).

GLib's macros are implemented with inline assembly, and probably a lot
faster than our naive bit shifting implementation.

The other audio_format attributes are not used.

This function is too complex for being inlined everywhere.

Send another "magic" MIME type when the byte order must be reversed.
This also fixes byte order issues when big-endian was involved (input
data or host byte order) - that was completely broken.

This function is not critical for performance, and the inline
expansion looks too expensive.

For better optimization.