feat(studio): add conversational DevEnv Skill workbench

[wujiaming-ai]

Summary

• add a chat-first Studio workflow for creating and optimizing Skills in dedicated DevEnv sessions
• support concurrent conversations, revision-scoped artifacts, preview, download, follow-up refinement, and publishing as a new or existing Skill
• make session lifecycle state-aware, including TTL expiry, released sandboxes, asynchronous snapshot recovery, and preserved completed artifacts
• harden Codex delegation with an unattended best-effort execution contract, untrusted-source boundaries, deterministic validation requirements, and an explicit artifact handoff protocol
• align Skill Workbench model credentials with Studio's Codex tooling and retain the existing public deployment URL during source deployments

Reliability and UX

• isolate follow-up request inputs from prior completed-task output fields
• bind downloads and publishing to immutable artifact revisions and SHA-256 digests
• recover artifact metadata without repeatedly downloading completed results
• prevent input after DevEnv expiry while guiding users to completed Skills
• expose provisioning, generation, validation, packaging, artifact download, recovery, and failure states with actionable messages
• enforce a 20 MiB upload limit and strict ZIP, path, UTF-8, file-count, expanded-size, and credential validation

Verification

• uv run pytest tests/cli/test_frontend_skill_workbench.py -q: 142 passed
• targeted VeFaaS upload regression: 3 passed
• uv run pytest -n 16: 1375 passed, 5 skipped
• frontend tests: 443 passed
• frontend TypeScript/Vite production build: passed
• uv run pre-commit run --show-diff-on-failure --color=always --all-files: passed

Deployment validation

• deployed version: 1.0.11.dev44+g0657d0084.d20260806
• deployment URL: https://sdr2plsgnd37e6e13ajb1.apigateway-cn-beijing.volceapi.com
• deployed frontend asset SHA-256 matches the committed build artifact