None of these commits changes any outputs in the regression tests. They simplify the cu-level QP adjustments to hopefully make them more robust, and I believe they fix issue #127 (where pmode + qg-size cause coding problems)