fix(responses): allow built-in tools in Responses requests

[stemirkhan]

Summary

• allow built-in Responses tools like image_generation, computer_use, and code_interpreter on Responses-family requests instead of rejecting them locally
• keep /v1/chat/completions strict so unsupported built-in tools still return invalid_request_error
• add focused unit/integration coverage for HTTP and websocket Responses flows plus an OpenSpec change

Testing

• /home/temirkhan/codex-lb-my/codex-lb-upstream/.venv/bin/python -m pytest tests/unit/test_openai_requests.py tests/integration/test_openai_compat_features.py tests/integration/test_proxy_websocket_responses.py -q
• openspec validate --specs
• manual local verification with the latest Codex for VS Code against a dedicated preview instance

Notes

• fixes Codex update has broken this #426
• intentionally narrower than fix(proxy): prefer budget-safe routing and support image-generation compatibility ("code":"invalid_request_error","param":"tools") #421: this only addresses the local tools validation regression for Responses-family clients