Feature: AI-assisted pre-labeling pipeline and UX enhancements

[YimingYang-MTU]

This PR introduces a AI-assisted labeling pipeline and several UX enhancements designed to accelerate 3D bounding box annotation for LiDAR point clouds.

The primary goal is to allow users to ingest model-generated proposals (e.g., from OpenPCDet, MMDetection3D) to serve as a baseline for manual correction, significantly reducing the "from-scratch" labeling time.

Key Features:

1. AI-Assisted Pre-Labeling: Integrated support for ingesting external 3D detection proposals to bootstrap the annotation process.
2. Spatial Copy-Paste Selection: Developed a new selection logic allowing users to duplicate and translate labels across sequential frames efficiently for one / more than one objects.
3. Enhanced Viewport Controls: Added automated viewport-locking for labeling processing stability and 180° orientation flipping to facilitate rapid inspection of object heading.
4. Allow editing of default object classes and corresponding default object size.