* implement A/52a downmix extensions
* use restrict pointers where appropriate
* reduce size of delay buffer by 50%

* get rid of globals in bitstream parsing
* get rid of statics in imdct.c
* avoid overflows, including reading the a52 stream !!!
* make dither code faster (generate dither table in advance ?)
* redo all bit allocation if previous frame had zero_snr_offsets
* make dynrng work in dual-channel streams

* update bitstream code