Speed up video analysis: cache face-detector net, lower default sampling fps

Merge upstream/main: reconcile async-DB/FastAPI/React fork with ML methodology branch

Replace 4-tier verdict system with a simple 60% AI/Authentic cutoff

solvje the CI error

feat: Add Wav2Vec2 audio detector and CLIP ViT-L/14 image detector

feat: Add production infrastructure, CI/CD, Docker, tests, and implementation plan

feat: Add v2 architecture — FastAPI + Gradio, core pipeline, CorefakeNet, training suite

Integrate audio deepfake detection module into multimodal pipeline

Update GradCAM visualization and video output

Initial commit: AI Authenticity Checker