fix: support Google Colab secrets for HF_TOKEN loading

[octo-patch]

Fixes #612

Problem

The Dummy Agent Library notebook code uses os.environ.get("HF_TOKEN") to read the token, which silently returns None in Google Colab. In Colab, secrets are stored in a separate keystore and must be accessed via google.colab.userdata.get(), not os.environ. This causes subsequent API calls to fail — but since the token assignment doesn't raise an error, users have no way to know what went wrong.

Solution

Updated the token-loading snippet in dummy-agent-library.mdx to:

1. Try google.colab.userdata.get('HF_TOKEN') first (Colab environment)
2. Fall back to os.environ.get("HF_TOKEN") in all other environments (local, Jupyter, etc.)
3. Explicitly pass token=HF_TOKEN to InferenceClient so the token is actually used

Testing

Documentation-only change to the code snippet shown in the course page. Logic is straightforward environment detection with a graceful fallback.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[sergiopaniego]

For the code parts of the course, we use https://huggingface.co/agents-course/notebooks/ as source. I've pushed the changes there

[sergiopaniego]

Thanks!!