AgenticEDA: Design the core Agentic loop on LangGraph

[indrayudd]

Here are some features of the loop:

• The agent is stateful. State is derived from the rules the agent is fed with.

• The agent's workspace is a .ipynb notebook (using nbclient).

• It should be able to read all types of outputs (text, dataframes, images) from cell outputs and choose to either take corrective steps or progress through the next state.

• We want to use Python and eventually use it with a JupyterLab backend.

FYI: @tkpratardan @gpsaggese