Tutorial coverage gaps: depth/3D, camera models, kornia.io, ONNX

An audit of the tutorial catalog (38 notebooks) found the coverage is heavily weighted toward augmentation (7 tutorials) and local-feature matching (5), while several major kornia modules have no tutorial at all:

- **Depth & 3D** — nothing covers `kornia.geometry.depth` (depth warping, unprojection) or point-cloud utilities.
- **Camera models** — `kornia.geometry.camera` (pinhole, projections, distortion) has no dedicated tutorial, even though it underpins the geometry stack.
- **Image I/O** — `kornia.io` / `kornia-rs` loading is only touched in passing (one category tag); a short tutorial on fast image loading (including `kornia_rs.io` decoding) would be a natural entry point. The new Ray VLM data-loading tutorial partially covers this.
- **ONNX / deployment** — no tutorial shows exporting kornia pipelines via `kornia.onnx` or running them outside PyTorch.

Smaller gaps worth considering: color constancy / white balance, and a geometry-solvers overview (`find_homography_dlt`, `find_fundamental`, RANSAC).

This came out of the same audit that merged several thin single-operator tutorials (#see recent commits on the audit branch); the intent here is to fill empty areas rather than add more depth where coverage is already dense.

Start by reviewing the 38-notebook catalog and the recent audit-branch commits to identify an empty area with a manageable scope. Use the relevant entry points named in the issue—kornia.geometry.depth, kornia.geometry.camera, kornia.io, kornia_rs.io, or kornia.onnx—and add a focused tutorial that fills one selected gap. Done means the chosen module has a clear, runnable notebook rather than another thin single-operator example.

Written by the indexing model from the issue text.