Conversation history: support R and Python workers without a Shiny session

I’m looking at the history work in #379 and wondering whether both the Python implementation and planned R port could support workers without an active Shiny session.

In Deputy, I’d like to let someone select a conversation branch in a Shiny app, continue it in a background R worker, and see the saved continuation when they return. I’d pass the user, chat and conversation identifiers to the worker from the app.

My immediate use case is R with Deputy and ellmer, but I’d like the same workflow to be available in Python with chatlas.

The separation of display history and provider turns looks useful here. I may summarize older turns for the model while keeping the original conversation available. I’d also like to save a Deputy run ID through the planned application-state support.

Would this fit the history redesign? I’d like to reuse the history operations and validation you’re building.

This also relates to #307’s conversation identity work and #61’s distinction between UI content and model turns. I’m tracking the Deputy integration in JamesHWade/deputy#66.

Start by reading the history redesign in #379, then compare the conversation identity work in #307 and the UI-content/model-turn distinction in #61. Review the Deputy integration tracking in JamesHWade/deputy#66 and determine whether reusable history operations, validation, and application-state support can cover workers without an active Shiny session; done requires an agreed scope for both R and Python.