Heavily improved noise fill, psychoacoustics, and some improvement to transient detection. Fixed a bug in the DS decoder that prevented noise fill from working correctly. Include a generic (x87, no SSE or AVX) encoder, and a batch script to convert using ffmpeg.