mason: show loading spinner for streaming responses

[annzhang-db]

Summary

• show an assistant loading bubble as soon as a generated chat UI starts a streaming request
• replace the spinner with the first streamed token and remove it on empty completion, interruption, or error
• keep the LangGraph and OpenAI templates aligned, including accessible text and reduced-motion behavior

Testing

• node --check for both generated UI scripts
• pytest tests/unit_tests/init_test.py -q (27 passed)
• git diff --check

The generated template route test was not run locally because the reused FastAPI/TestClient environment hangs on GET / before reaching the new static-asset assertions.