and not the actualy quantization after possible incrmements to
avoid saturation.
Experimental: fixed some obvious oops in the look-ahead rate-controller
adjusted a ridiculous large gain in pass-1 and put in
some additional 'sanity' limits on rate-control feedback.
