feat: Automatic scene complexity-based encoding quality adaptation

[wmehanna]

Problem

Currently, BitBonsai applies uniform encoding quality (CRF) across entire videos. This is inefficient because different scenes have vastly different encoding complexity:

• Simple scenes (static content, low detail): Need fewer bits — current approach wastes bits here
• Complex scenes (high motion, grain, high detail): Need more bits to maintain quality — current approach may under-encode here

Proposed Solution

Implement automatic scene complexity analysis before/during encoding that:

1. Detect scene changes using FFmpeg's scene detection filter to build a scene complexity map
2. Classify scene difficulty based on spatial/temporal complexity metrics
3. Adapt CRF per-segment: Apply lower CRF (higher quality) to complex scenes, higher CRF (lower quality, more compression) to simple scenes — all while maintaining overall bitrate efficiency
4. Use FFmpeg's per-title encoding pattern: Generate weighted encoding complexity analysis that informs the encoding pass

This is transparent to users — no configuration needed.

Why This Matters

• Better quality at same bitrate: Bits allocated where eyes are most sensitive
• Better compression at same quality: Simple scenes compressed aggressively without perceptual loss
• Industry-standard technique: Netflix and others use per-scene complexity analysis extensively
• Complements existing VMAF validation: After encoding, VMAF already validates quality; this makes the encoding itself smarter

Technical Approach

In FfmpegService:

1. Add analyzeSceneComplexity(inputPath) method using scene detection filter
2. Build complexity profile: segments with complexity scores (0-1)
3. Map complexity to CRF adjustments: baseCRF + (1 - complexity) * crfDelta
4. Use FFmpeg segment filtering or two-pass approach to apply per-segment CRF

Example complexity classification:

• Complexity 0.0-0.3: Static title screens, fade transitions — CRF +4 (aggressive compression)
• Complexity 0.3-0.7: Normal content — CRF base (no change)
• Complexity 0.7-1.0: Action sequences, grainy film, CGI — CRF -4 (protect quality)

Labels

enhancement, video-encoding