Add Keypoints message type and expand dai::ImgDetection with dai::RotatedRect and dai::Keypoints

[aljazkonec1]

Purpose

This PR is the first port of depthai-nodes ImgDetectionsExtended (link) and Keypoints (link) messages into the core library.
An example of using new ImgDetections + Keypoints can be found on this branch.

Specification

• New feature dai::RotatedRect: Added two new utility functions to dai::RotatedRect, used for converting the stored rectangle to the other popular YOLO and VOC bounding box representation types.
• New Keypoint common data structure: It is defined as a collection of Point3f coordinates, float confidence, int label and string labelName.
• New KeypointsList common data structure: It stores a vector of dai::Keypoint objects, a vector of possible edge idices between Keypoints and provides utility functions for faster and easier conversions between dai::Keypoints types and other common types like dai::Point3f / dai::Point2f.
• New Keypoints message type: serialization of dai::KeypointsList.
• Extended dai::ImgDetection :
  • Add dai::RotatedRect as a field
  • Add dai::KeypointsList as field
  • Add utility functions and getters for new fields and easier access to relevant data
  • If using getters / setters, the legacy xmin, ymin, xmax, ymax format and new dai::RotatedRect fields are synced and current implementations of ImgDetections still work.
• Pybindings for all new additions.

Dependencies & Potential Impact

Potential risks include already built pipelines that rely on dai::ImgDetections.

Testing & Validation

Tested on this branch. Two annotations are seen. detections topic sends dai::ImgDetections directly to visualizer and visualizer handles annotating it. modified_annotations is a modified version where rotation and keypoints are also taken into account when visualizing a bounding box. Tested only on RVC4.