[Tremor] fixed_point_encoder

[Timothy B. Terriberry]

Ralph Giles wrote:
> over the analysis and mode choice code from libvorbisenc. The main issue
> is to ensure you maintain enough dynamic range during calculations.

The good news is that because you're writing an encoder, not a decoder, 
you don't have to support everything (in particular, floor0, which is 
the hardest part to do with enough precision). So in that sense encoding 
is much easier than decoding.