Added an API and docker distribution for building production ready application

[Mawwlle]

Add production API, web demo, and Docker support

What was done

API

• Added CORS middleware to allow browser requests
• Frame extraction moved from upload time to session creation (lazy, cached)
• MOV support — non-MP4 files are converted to MP4 via ffmpeg on upload
• Session close now deletes uploaded files and frames from disk
• Added /videos/upload-frames endpoint for uploading raw frame sequences
• frame_count added to video metadata via ffprobe nb_frames

Bug fixes

• Fixed MPS crash during propagation: torch.tensor() was inferring float64 from Python floats in Kalman filter IOU computation (sam2_base.py:472); forced dtype=torch.float32

Infrastructure

• Added run.sh — handles PYTHONPATH and SAMURAI_APP_ROOT so uvicorn launches correctly without the sam2/ directory shadowing the package
• Added Dockerfile with CUDA 12.1, ffmpeg, checkpoint download at build time
• Added docker-compose.yml for single-command GPU deployment on port 7676; Docker-specific paths (SAMURAI_APP_ROOT, SAMURAI_DATA_PATH) set via environment: to override .env
• Added .env.example

Frontend

• Added index.html — single-page demo with drag & drop upload, canvas bbox drawing, propagation streaming, mask overlay rendering, and video export with annotations

Docs

• Updated README.md with native run instructions, Docker/Compose instructions, endpoint reference, and typical workflow

[sourcelocation]

LGTM