- Previous fix broke dependent  quantization for the non-scaling list case.
  - Negative shift values are no longer set to zero, so the shift is applied repeatedly to the fixed invQScale value (g_invQuantScales) during each loop.