Switch to fixed-point algorithm

[GlaireDaggers]

Currently with larger keyframe intervals (even at just 60 frames) accumulating errors are very noticeable and result in a "pulsing" behavior.

Will likely need to switch to a fixed-point integer algorithm instead, so that any error is 100% predictable can be precisely accounted for in the encoder.

[GlaireDaggers]

Implemented as of b1d2679
I figure this also warrants a minor codec version bump since accumulated errors would likely be exacerbated by trying to run new fixed-point output through a floating-point decoder.