Summary

• request both messages-tuple and values stream modes for chat runs so reasoning-only chunks can reach the UI incrementally
• add frontend coverage for reasoning-only AI messages
• add backend serialization coverage to ensure reasoning_content on AIMessageChunk is preserved

Closes #2785

Tests

• uv run pytest tests/test_serialization.py -q
• uv run ruff check tests/test_serialization.py
• pnpm exec vitest run tests/unit/core/messages/utils.test.ts
• pnpm exec tsc --noEmit