Updating v3 docs post JupyterCon

[srdas]

This is an incomplete draft of a revision of all the documents for Jupyter AI for the v3 version, which launched in beta at Jupyter Con 2025. (WIP, no review required at this time.)

Notes (WIP), other than the changes to the docs itself:

1. Removed all discussion of pydantic as it is not relevant any more, especially in the user section.
2. The model provider table is not needed now with LiteLLM as the list is huge. Some of the API_KEYs are changed also. Removed the table as it is best to point to the documentation and let people look it up.
3. Commented out conda installation for now.
4. Customization of the chat interface with a config.py file does not work any more so I have commented it out.
5. Since OpenRouter is handled easily by LiteLLM, deleted the separate page on using OpenRouter, this is also not supported any more.
6. Amended the Bedrock usage page for the latest screenshots and commented out the parts that deal with accessing Bedrock models with ARNs. Also, it is not clear how we will handle fine-tuned models with an ARN, as it is not supported but there seems to be no need for this feature, so I commented out this documentation.
7. SageMaker endpoints usage section: Also commented this out as it does not seem to be required as we do not support this any more. Will this be required when SageMaker is upgraded to use v3? Maybe best to have a separate persona for working with SM endpoints?
8. GPT4All - commented this out as well as it seems to be unnecessary. All these models are rather old and the newer ones are on Ollama, for which we have a section explaining its use.
9. Removed the doc for vLLM as it is now handled via LiteLLM.
10. Using magics with SageMaker endpoints has been deleted because this is no longer supported by our implementation. This may need to be included later.
11. Removed config for embedding and completion models as these are no longer used in the same way.