[codex] Add ChatGPT MCP deployment preflight

[yha9806]

What changed

• Added deploy/chatgpt-mcp/ for a direct Cloud Run deployment of the ChatGPT review MCP endpoint.
• The deployment image installs the published vulca[mcp]==0.23.1 PyPI package and runs vulca-mcp-remote.
• Added scripts/chatgpt_app_preflight.py to verify the submission JSON, remote MCP tool list, tool annotations, output schemas, test case coverage, and forbidden response metadata.
• Added resubmission evidence docs and updated the checklist to use a direct Cloud Run URL rather than vulcaart.art/mcp.

Why

OpenAI rejected prior submissions for incomplete privacy/screenshot evidence. This PR makes the production MCP deployment and resubmission preflight explicit, repeatable, and limited to the five review-safe tools.

Validation

• git diff --check
• python -c "import yaml; yaml.safe_load(open('deploy/chatgpt-mcp/cloudbuild.yaml'))"
• PYTHONPATH=src /opt/homebrew/bin/python3 scripts/chatgpt_app_preflight.py --submission chatgpt-app-submission.json
• PYTHONPATH=src /opt/homebrew/bin/python3 -m pytest tests/test_mcp_remote_profile.py tests/test_visual_discovery_docs_truth.py -q -> 27 passed

Remaining before OpenAI resubmission

• Build and deploy the Cloud Run service from an environment with Docker/gcloud or Cloud Build access.
• Run the preflight script against https://<cloud-run-service-url>/mcp.
• Configure that URL in the OpenAI Platform app dashboard.
• Capture fresh ChatGPT app screenshots from the production MCP endpoint.

Docker and gcloud were not available in the local Codex environment, so container build and production deploy were not executed here.