Batch cubemap splitter with mask support and per-face XMP rig sidecars

[Create-your63]

Motivation

• Convert the single-image equirect->cubemap utility into a folder-based batch tool so already-extracted frames can be processed in bulk.
• Ensure mask images (already produced by the user's extractor) are projected the same way and emitted with filename templates that match RealityCapture/RealityScan pairing conventions.

Description

• Added tools/equirect_to_cubemap_xmp.py, a folder-oriented tool that finds equirectangular source images and for each source emits six cubemap face images, six .xmp sidecars and optional six mask faces using a paired --mask-dir lookup.
• New CLI options: --mask-dir, --mask-suffix, --image-name-template, --mask-name-template, --face-size, --image-format, --jpg-quality, and --xmp-namespace to control naming, mask pairing and output format; image naming uses the provided templates via render_name.
• Image sampling uses bilinear reconstruction and mask sampling uses nearest-neighbor to preserve mask regions; each face receives an XMP with xcr:Rig, xcr:RigInstance, xcr:RigPoseIndex, and a rotation matrix; all per-frame results are collected into cubemap_rig_manifest.json.
• Updated README.md with a usage example showing the folder+mask workflow and the example command python3 tools/equirect_to_cubemap_xmp.py ./eq_frames ./cubemap_out --mask-dir ./eq_masks ....

Testing

• python3 -m py_compile tools/equirect_to_cubemap_xmp.py succeeded (syntax check passed).
• Attempted to run python3 tools/equirect_to_cubemap_xmp.py --help in this environment failed with ModuleNotFoundError: No module named 'PIL' because Pillow is not installed here (runtime dependency), so full runtime execution was not validated in this environment.

---

Codex Task