feat(chat): display expandable reasoning and activity traces - #4

Summary

• display an expandable “Thinking” panel for each assistant turn, streaming the reasoning content actually returned by the model API through reasoning_content or reasoning
• include real search, browsing, and tool-execution events in the same activity trace without generating or fabricating hidden chain-of-thought
• keep the trace expanded while the assistant is working, then collapse it into a “Thought for N seconds” summary that users can reopen
• persist activity entries with conversations and represent completed, stopped, and failed states

Validation

• npx tsc --noEmit -p tsconfig.json
• desktop Open UI production build
• desktop Open service-worker build, including verification that every manifest-referenced file exists
• reloaded the unpacked extension in Edge and manually verified running, completed, expanded, and collapsed states