Update example for converting DLC project to benchmark

[niksirbi]

Description

What is this PR

• Bug fix
• Addition of a new feature
• Other

Why is this PR needed?

We need a working end-to-end example showing how to convert a DeepLabCut (DLC) project to the poseinterface benchmark dataset format.
This is both for us to actually use when converting DLC projects, and to provide a reference example to collaborators.

What does this PR do?

Warning

This is a Frankenstein branch pulling code from multiple ongoing PRs, including #26 and #39 and #45.
I created this to facilitate quick-and-dirty conversion. It will need to be rebased to main and cleaned up after these other PRs have been merged.

• Adds a lightweight DLC project fixture under tests/data/dlc/ with 2 sessions, each with a small (100 frames) video, placeholder PNGs (0-byte test files), annotation CSVs, and prediction CSVs (first 100 frames, corresponding to the videos).
• Adds the convert_dlc_to_benchmark sphinx-gallery example that runs against the fixture. This example:
  • Uses annotations_to_poseinterface (from Refactor annotations_to_coco #26) for converting DLC annotations to COCO keypoints format.
  • Adds a frames_to_poseinterface utility to copy and rename frame images according to filenames in a COCO .json file.
  • Uses video_to_poseinterface (from Format videos to poseinterface spec. Extract clip function.  #39) for re-encoding and renaming videos to our format.
  • Uses the predictions_to_poseinterface function from Convert predictions to cliplabels.json using movement #45 for converting DLC predictions csv to cliplabels.json format.
  • Uses extract_clip (from Format videos to poseinterface spec. Extract clip function.  #39) to extract clips from the session video and the corresponding cliplabels.json file, and save them in the Clips/ directory of the benchmark dataset
• Adds a tree utility function to display directory structures, with trailing / on directory names. This is called several times in the updated sphinx-gallery example to show the structure of the DLC project and the converted benchmark dataset.
• Updates docs config to support sphinx-gallery execution (resolves nbsphinx conflicts, adds matplotlib dependency).

This is a rough schematic of the workflow illustrated by the example:

What is currently missing?

Waiting for the aforementioned 3 PRs to be reviewed and merged, so this branch can be rebased.

References

• Refactor annotations_to_coco #26
• Format videos to poseinterface spec. Extract clip function.  #39
• Implement a predictions_to_coco function #38
• Convert predictions to cliplabels.json using movement #45
• Utility for extracting a clip from a video #10

How has this PR been tested?

• The sphinx-gallery example runs end-to-end against the test fixture (python examples/convert_dlc_to_benchmark.py).
• The docs build successfully with make clean html from docs/.
• Unit tests have been added for utils.tree and for io.frames_to_poseinterface.

Is this a breaking change?

No.

Does this PR require an update to the documentation?

Yes. The new sphinx-gallery example (convert_dlc_to_benchmark) is automatically rendered into the documentation gallery. The docs config has been updated to support this.

The new tree utility has been added to the API reference.

Checklist:

• The code has been tested locally
• Tests have been added to cover all new functionality
• The documentation has been updated to reflect any changes
• The code has been formatted with pre-commit

[niksirbi]

Closing this as superseded by #49