Layer-1 support not heavily tested .. 
i-stereo in Level-3 seems to work now, but I have to optimize it.
l3: exchange the DCT in III_Hybrid by a faster version.
general optimizations in Level-3 code.
add a few more audio-systems.
add CRC check.
no 'free format stream' support.
in layer2.c:
	step_two: fraction as pointer .. 
	process first channel 0 than channel 1
	copy channel 0 to channel 1 for: i >= jsbound

- MAYBE it's possible to save half of the muls in SubBandSynth .. because
  they are duplicated. In this case we need a big tmp table
  (about 256 * 512 'real' values) (I check this .. doesn't seem to be possible!)

- better frame reader .. especially for layer 3 

- check again layer1 !!

- integrate sign-check into huftables ... at least for hufman_count1 ...
- Put the huffman decoding into the dequantize part 

- put decode_4to1, decode_2to1 and normal decode together to
  make it selectable on the command line. Write a special 'single channel'
  decode.


