draft(pky-py): Use extended task

[gadenbuie]

This doesn't work yet, but I haven't been able to figure out why. Pairs with #63.

To demo, use pkg-py/examples/app.py and ask a question like

Show only the passengers who survived and then give me a short history of the Titanic.

and note that the dataframe doesn't update until the LLM response is complete. Ideally, the data update should be more-or-less immediate and should finish before the chat streaming is done.