Adds image extracted features observation term and cartpole examples for it

[glvov-bdai]

Description

This adds an observation term to be able to easily extract features from the images, and adds a cartpole example of using this new term.

The new ResNet18 cartpole converges in less than 100 epochs.

Type of change

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

Checklist

• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have updated the changelog and the corresponding version in the extension's config/extension.toml file
• I have added my name to the CONTRIBUTORS.md or my name already exists there

I will update the version in the changelog and extension.toml after approval prior to merging in due to it causing merge conflicts when main updates

[Dhoeller19]

Lgtm, thanks! Please update the changelog

[jsmith-bdai]

Overall looks good, thanks for adding! Just left some minor comments.

Just curious - how do you ensure the models are frozen? I didn't see it configured anywhere

[glvov-bdai]

@jsmith-bdai Feature extraction encoders are frozen due to the .eval() flag in the model creation, like in

lambda: getattr(models, resnet_name)(pretrained=True).eval().to("cuda:0"),

Anyways, even if they weren't in eval mode I think they would still be frozen as I don't back-propogate the loss anywhere with model so weights shouldn't update