Update assistant timestamp on first token Make chat message timestamps observable so the UI updates when an assistant response starts streaming visible text. This sets the assistant timestamp when the first non-thinking content arrives and clarifies the Foundry Local thinking-filter TODO.