Decide agent framework to use

[dlqqq]

We will need to add agentic tool-calling capabilities to Jupyternaut by v3.0.

I've explored using Pocketflow in this PR, prior to the codebase split: jupyterlab/jupyter-ai#1479

Piyush noted that we may be able to simplify the flow here by leveraging a more fully-featured agent framework. Notably, Agno and Langgraph are well-maintained and see much higher usage than Pocketflow. These options may prove to be better suited to our needs in the long term.