[codex] Fix Visora EgoForce sync and diagnostics

[gaurangbharti1]

Summary

• Fix Visora video processing to select frames from ARKit timestamp rows and read those exact frame indices, avoiding RGB/ARKit/LiDAR desync on VFR MP4s.
• Disable stale YOLO detector tracking/reuse by default for frame-level EgoForce inference.
• Add browser-playable Visora runner, jitter analyzer, and pipeline visualization scripts for step-by-step debugging.
• Make inference imports and RSS lifting more robust in the local environment.

Root Cause

The problematic Visora MP4 could not be reliably processed with naive OpenCV sequential frame reads after seeking. RGB frames, ARKit rows, and LiDAR depth could drift out of sync, making EgoForce outputs look explosively unstable.

Validation

• python -m py_compile camera_models/__init__.py core/rss.py demo/demo_utils.py demo/inference.py scripts/run_visora_video.py scripts/analyze_visora_jitter.py scripts/visualize_egoforce_pipeline.py
• Ran row-synced jitter analysis and generated H.264 overlay/pipeline MP4s locally.