Smoother conversation and book loading

Switch between recent conversations without a blank transcript, and open books and chapter lists from the local world snapshot while fresh data loads.

Recently opened conversations stay visible while the assistant reconnects. This fixes a pause where the cached transcript could appear, disappear, and then return. The message box becomes available after the connection has loaded the current history and saved model choice.

Books and chapter lists can appear from the current account’s local world snapshot while fresh data loads. The main page also loads separately from the assistant’s heavier display code.

No action is required beyond loading the updated app. Conversation previews remain temporary and account-specific; a full reload still reconnects to the server. These changes do not alter the time a model takes to generate a response.

The local Unfinished Atlas demo below switches between Mara’s and Iona’s conversations. The recent transcript stays visible while the message box waits for the current connection. The demonstration includes an artificial connection delay; it is not a production speed benchmark.