Refactor and improve the Camera Sensor and example

[edbeeching]

No description provided.

[Ivan-267]

This is currently in progress: edbeeching/godot_rl_agents_examples#14 (comment)

This should fix the sensor for Godot 4.2 so it works during training.

What will still be left is to look into:

• Onnx export / inference support (I didn't yet test whether this works)
• Add downsampling option for rendered images before sending as obs
• Potential grayscale feature (reduce the size where colors are not necessary)
• Maybe having an option to add depth information as the 4th channel

As well as any other features that might be useful.

[edbeeching]

Hey @Ivan-267 are any items left to do on this?

[Ivan-267]

Those were just some proposed options, none are implemented yet. I've tried the sensor a bit after writing those and I can confirm that we can't export onnx / run inference right now.

Edit: I added the idea about downsampling from discord here as well. The list is sorted by importance, to fully support the sensor we need onnx export, other enhancements are optional.