+ libmad_input now takes one frame at a time (see mad_adec.h for defines)
+ audio_linear_dither() gives better sound quality then s24_to_s16_pcm(),
  but also makes audio artefacts louder so disabled it for now (see file
  mad_libmad.c)
+ cleaned up libmad_output (see file mad_libmad.c)
+ cleaned up InitThread (see file mad_adec.c)
+ Made buffer size match libmad's expectations (see file mad_adec.h)
+ updated documentation