Feature/docker nuscenes

[hakuturu583]

Summary

• Add Buildx bake targets for reproducible images: base (CUDA 11.7, core deps, gsplat, pytorch3d, nvdiffrast, smplx, PYTHONPATH preset) and nuscenes (adds nuscenes-devkit plus isolated SegFormer conda env)
• Provide Docker usage docs (docker/README.md, README, NuScenes guide) including TAG_PREFIX override and sample run with data mount
• Add scripts/nuscenes.sh to run the entire NuScenes preprocessing pipeline inside the container: optional CPU/GPU, split/scene selection, interpolation/workers, dynamic mask toggle, SegFormer checkpoint download, and optional preprocessed human-pose download
• Keep scripts/nuscenes.sh tracked via .gitignore tweak and use python3 inside the container commands to align with installed deps

Usage notes

• Build images: docker buildx bake (produces drivestudio:base and drivestudio:nuscenes; override tags with TAG_PREFIX=...)
• Run preprocessing: ./scripts/nuscenes.sh /abs/path/to/nuscenes/raw --split v1.0-mini --num-scenes 10 --interpolate 4 --workers 32 (will build the image if missing, mount repo+raw data, run preprocessing, extract sky/fine dynamic masks via SegFormer, and optionally download the SegFormer checkpoint and preprocessed humanpose archive)

Testing

• Not run (container build and preprocessing not executed in this environment)