Update OpenCV to 4.18.X and allow numpy 2.X

[NiklasNeugebauer]

Motivation

We are still using numpy 1.X. This is because we were stuck at on opencv 4.8 since newer versions at the time had issues with the omnidir camera calibration module. This seems to have been fixed and as such we want to allow newer numpy versions and a new opencv minor version.

Implementation

• The numpy dependency is now <3.0.
• The opencv-python-contrib dependency is now >=4.12.0.88 (current newest) and <4.13.0.0.
• opencv-python was removed (not needed with opencv-python-contrib)
• cv2 is ignored by mypy (installing stubs takes more than 5 minutes)
• Tests for real hardware cameras were fixed (likely unrelated, but was not a big fix)

Progress

• I chose a meaningful title that completes the sentence: "If applied, this PR will..."
• I chose meaningful labels (if GitHub allows me to so).
• The implementation is complete.
• Pytests have been added (or are not necessary).
• Documentation has been added (or is not necessary).

[NiklasNeugebauer]

There seem to be some issues with opencv`s jpeg encode/ decode. Let's wait for #356 where we use PIL and TurboJPEG instead.

[falkoschindler]

Checks are still failing 🫤

[NiklasNeugebauer]

yes...

[falkoschindler]

Ok, some of the newly required type annotations feel a bit redundand, but I guess we can't do much about it.

In d4e292e I made just some minor changes:

• improved the computation of percentiles to get all five values in one pass,
• refactored the instantiation of intrinsics to reduce the diff and avoid # type: ignore,
• wrapped a very long line,
• reverted decoded = decode_jpeg_image(...) by specifying a more precise return type ImageArray, and
• using ImageArray in the signature or from_array.

Ready to merge if you agree. 👍🏻