fix(studio): stabilize parallel agent rendering

[evanlowe]

Summary

• Aggregate streamed events by author and invocation so parallel outputs remain independent and repeated loop turns remain distinct
• Bind event IDs only to final persisted events and suppress empty or control-only transcript cards
• Present parallel agent output responsively: stacked on narrow screens, up to three columns with horizontal overflow on wide screens
• Add a Studio workflow matrix covering LLM, Sequential, Parallel, Loop, mixed, and all-agent compositions
• Rebuild the bundled Studio frontend assets

Testing

• Frontend test suite: 1040 passed
• Frontend production build passed
• Workflow matrix Python tests: 11 passed
• Ruff checks, formatting, and secret scanning passed
• Studio end-to-end scenarios passed on Volcengine and BytePlus for individual and composed agent types