Add batch cubemap splitter with mask projection and RealityCapture-compatible naming

[Create-your63]

Motivation

• Provide a focused tool that takes already-extracted equirectangular images (and optional mask images) and splits each into six cubemap faces while grouping those faces as one fixed rig for RealityCapture/RealityScan imports.
• Support the user's workflow where image/frame extraction and mask creation are done externally, so this tool only handles splitting, mask projection, XMP sidecars and manifest generation.

Description

• Added tools/equirect_to_cubemap_xmp.py which supports single-file and folder batch processing, optional mask handling via --with-masks, and RealityCapture-compatible mask naming modes (--output-mask-mode with dot-mask and suffix).
• The script finds input masks by trying stem + input_mask_suffix (default _mask) and name + .mask.<ext>, projects RGB with bilinear sampling and masks with nearest-neighbor, writes per-face images and .xmp sidecars containing rig and rig-instance UUIDs and a camera-to-world rotation matrix, and emits cubemap_rig_manifest.json summarizing results.
• Added IMAGE_EXTENSIONS whitelist and list_input_images() to gather input images while skipping mask files, and --base-name / --face-size / --image-format / --jpg-quality / --xmp-namespace options to control outputs.
• Updated README.md with usage examples, mask lookup rules, output naming examples, and workflow notes recommending png and explaining mask projection behavior.

Testing

• python3 -m py_compile tools/equirect_to_cubemap_xmp.py succeeded (syntax check passed).
• Invoking python3 tools/equirect_to_cubemap_xmp.py --help failed at runtime due to a missing numpy dependency in this environment (ModuleNotFoundError).
• Attempt to install runtime deps with python3 -m pip install --quiet numpy pillow failed due to network/proxy restrictions (403), preventing runtime verification of processing steps.

---

Codex Task