Add direct OpenAI Responses provider

[cmpadden]

Summary

• add an explicit OpenAI provider using OPENAI_API_KEY and OPENAI_BASE_URL
• share bounded Responses HTTP and SSE handling with the existing Codex transport
• support provider selection with configured OpenAI-compatible model IDs while keeping Gateway and subscription credentials isolated
• document direct OpenAI and compatible endpoint setup

Verification

• zig fmt --check src/
• zig build
• zig build test
• ./benchmarks/startup.sh --quick
• exercised ./zig-out/bin/fx provider openai
• exercised ./zig-out/bin/fx ask --json end to end against a loopback Responses SSE server, including bearer authentication, model selection, tools, streamed output, and clean stderr

Notes

• OpenAI-compatible model discovery is intentionally not assumed; configure FX_MODEL or models.openai.
• Full CI and the final ship gate are still required before this draft is marked ready.

[vercel-security-reviewer]

Security review: https://reviewer.vercel-security.com/vercel-labs/fx/pull/543