feat(studio): generate Agent configurations from natural language

[evanlowe]

Summary

• add an Ark output-schema planner that converts natural-language requirements into validated recursive Agent drafts
• add the Studio smart-generation input with loading, success, retry, and model-error states
• preserve nested Agent models, tools, memory, knowledgebase, and tracing settings when applying generated drafts
• add backend and frontend coverage for complex video, image, and operations workflows

Tests

• npm test — 233 passed
• npm run build — passed
• uv run pytest -q tests/cli/test_generated_agent_planner.py tests/cli/test_generated_agent_backend_codegen_extended.py — 39 passed, 3 skipped
• uvx ruff check veadk/cli/generated_agent_planner.py tests/cli/test_generated_agent_planner.py — passed
• uv run pytest -n 16 — 936 passed, 7 skipped; 4 failures and 1 collection error are limited to the local Python 3.14 environment missing optional llama_index and greenlet dependencies. CI installs all extras on Python 3.10/3.12.