Fixed: overlay too aggressive on media (regression from v4.9)
The v4.9 segment-fill was injecting a fixed, brightness-driven intensity: any bright + roughly-static region instantly snapped to a near-full overlay sheet, bypassing the slow accumulation that media is supposed to get. Reworked so segmentation decides where and accumulated risk decides how strong:
- Segments are now detected by content coherence (uniform regions, any brightness) instead of brightness.
- Fill level is the region's mean accumulated risk - dynamic, not a fixed value, so it tracks real protection need.
- Moving content self-suppresses (risk is already motion-gated), so video no longer reads as instant panels.
Net effect: overlay on media stays quiet and ramps gradually like it should to detect static logos,banners,etc on videos, while still flattening genuine static regions evenly (no patchiness).