Allow arbitrary camera frame dimensions

[alexander-soare]

What this does

Allows arbitrary camera dimensions for OpenCVCamera. Previously, if the camera dimensions were not supported by the camera, an exception would be raised. With this PR, the frame is resized as a postprocessing step.

How it was tested

Added test which has full coverage if you have a camera connected, but only partial coverage otherwise (see the test docstring).

I also tested it locally with a camera.

How to checkout & try? (for the reviewer)

Run this script.

import cv2

from lerobot.common.robot_devices.cameras.opencv import OpenCVCamera, OpenCVCameraConfig
from lerobot.common.utils.utils import init_logging

init_logging()

height = 40
width = 40
camera = OpenCVCamera(0, OpenCVCameraConfig(height=height, width=width))
camera.connect()
assert camera.height == height
assert camera.width == width
img = camera.read()
cv2.imshow("window", img)
cv2.waitKey(0)

You should probably get logs like:

INFO 2024-10-03 14:48:44 s/opencv.py:286 The provided width (40) could not be set natively on your camera. The settings are now (width=160, height=90. This class will resize the frames to your desired resolution before returning them.
INFO 2024-10-03 14:48:44 s/opencv.py:288 The provided height (40) could not be set natively on your camera. The settings are now (width=160, height=90. This class will resize the frames to your desired resolution before returning them.

You should see a window with 40x40 image appear.

[Cadene]

Looks good to me. Could we do the same for intealrealsense.py after merging/rebasing master?
Then I am happy to quickly review again and approve

[Cadene]

After rebasing on top of master, could we add the mock version to it?