LiDAR detection

[tejasstanley]

Summary

This PR adds the initial WIP structure for LiDAR detection support in PerceptionMetrics.

Changes

• Added a LiDAR detection dataset wrapper for PyTorch models.
• Added a placeholder TorchLiDARDetectionModel class.
• Added a read_points() interface to the LiDAR detection dataset base class.
• Added an initial nuScenes LiDAR detection dataset adapter.
• Reads nuScenes point clouds as x, y, z, intensity.
• Uses nuScenes sensor-frame boxes in [x, y, z, dx, dy, dz, yaw] format.

TODO

• Wire LiDAR detection inference with a backend such as MMDetection3D.
• Finalize the model input pipeline for point cloud detection.
• Decide and implement 3D detection metrics.
• Add evaluation support for LiDAR detection models.
• Add prediction export format for 3D boxes.