feat: Add global and per-class bounding box visibility toggles with filtering

[axthur]

Summary

When labeling complex datasets, users often encounter cluttered images with numerous overlapping bounding boxes. Additionally, reviewing or refining a single, poorly predicted class from pseudo-labeling can be tedious when all other classes are displayed.

This Pull Request introduces a comprehensive visualization filtering system, allowing users to toggle the visibility of all bounding boxes globally or filter specific classes individually. It also ensures that core actions (like clearing data) respect these visibility filters.

Key Features

• Global Bounding Box Toggle: Added a "Visualize Bounded Boxes" checkbox next to the existing class name toggle. It can be triggered via the B keyboard shortcut to completely hide/show all bounding boxes, making it easy to inspect the raw image.
• Per-Class Visibility Toggles: Integrated checkmarks into each row of the classes table widget on the right. Users can now check/uncheck individual classes to isolate them for focused review or adjustment.
• Visibility-Aware Actions: Destructive or bulk operations now respect the visibility filtering. For example, using Ctrl + Delete or Ctrl + Backspace will only delete visible boxes, leaving hidden classes untouched.

Other

• Checkbox Startup Bugfix: Resolved a UI desynchronization bug where the "Visualize Class Name" checkbox started visually unchecked on startup despite class names being drawn on screen (which previously required two clicks to sync).
• Alternative Shortcut: Added Ctrl + Backspace as an alternative shortcut alongside Ctrl + Delete for clearing label data, resolving shortcut collision issues.
• New Shortcut: Pressing B hides/shows all bounding boxes.