chore(version): v19.5.0

[janickm]

This pull request updates the project to version 19.5.0 and introduces support for the Argoverse 2 dataset in the NCore V4 converter, along with several bug fixes, optimizations, and documentation improvements. The most important changes are summarized below.

Major Feature Addition:

• Added support for the Argoverse 2 Sensor Dataset in the NCore V4 converter, including all 9 cameras, two Velodyne VLP-32C lidar units, and cuboid labels. This is the first dataset with a synchronized stereo camera pair and improved lidar modeling.

Bug Fixes and Improvements:

• Updated Bazel dependencies and exposed CameraModelParameters in the public API.
• Memoized EncodedImageData.get_decoded_image per instance to improve performance.
• Fixed flaky quantized depth roundtrip tests.

Documentation:

• Improved documentation formatting for better readability.

Versioning:

• Updated the embedded version label in .bazelrc to 19.5.0.