Minor follow-up refactors for DeepLabCut integration

[deruyter92]

Minor follow-up refactors on PR #19, for better integration with DeepLabCut.

• Accept typed Pose2DResult as valid input for pose_3d() method, instead of separate keypoint array and image size. This is a natural addition, since the result of prepare_2d() can directly be used in pose_3d() without unpacking. Providing unpacked keypoint array and image size is also kept as valid signature to flexibly support users that bring their own 2d keypoints.
• Move 3D pose input validation using valid_frame_mask: instead of performing validation and masking of valid frames between prepare_2d() and pose_3d() steps in the predict() method, all validation logic is moved to pose_3d(). This means that predict(...) method becomes literally equivalent to pose_3d(prepare_2d(...).
• An annoying (frame-by-frame) tqdm progress-bar is silenced.

[Copilot]

Pull request overview

This pull request refactors the DeepLabCut integration to improve the API design and make the pose estimation pipeline more composable. The changes enable users to pass Pose2DResult objects directly to pose_3d() instead of unpacking them, and consolidate all 2D result validation logic into the pose_3d() method rather than splitting it between predict() and pose_3d().

Changes:

• Accept Pose2DResult as input to pose_3d() method in addition to raw keypoint arrays, making the API more ergonomic
• Move 3D pose validation and frame masking logic from predict() into pose_3d() for better separation of concerns
• Silence frame-by-frame progress bar in DeepLabCut's superanimal_analyze_images call

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[xiu-cs]

Hi @deruyter92, this PR looks good to me. Please feel free to merge it.