feat(studio): support multimodal conversations

[yaozheng-fang]

Summary

• add multimodal upload, model response, and history support for images, TXT/Markdown, PDF, and video
• store media in local /tmp storage by default or tenant-isolated TOS prefixes, and resolve media safely for remote AgentKit runtimes
• add compact media previews, image lightbox support, / skill selection, and @ sub-agent targeting
• include a multimodal demo agent, default PDF rendering dependencies, documentation, and focused regression coverage

Validation

• uv run pytest tests/multimodal tests/cli/test_frontend_invocation.py tests/cli/test_frontend_trace.py tests/cli/test_frontend_runtime_proxy.py -q (35 passed)
• Ruff and Pyright checks passed for the changed Python modules and tests
• npm run build
• Markdown lint and git diff --check