feat(eval): add missing minimax backend configuration

[Tanmay9223]

🎯 What: Adds missing configuration setup in scripts/eval_only.py to properly support the minimax_chat backend.
💡 Why: skillopt-eval was failing to set up external backend calls for minimax correctly because the invocation to configure_minimax_chat was entirely missing, unlike in skillopt-train. This covers the edge cases not addressed by PR #85.
✅ Verification: Verified locally that running scripts/eval_only.py with backend: minimax parses arguments and configures the environment without crashing.
✨ Result: Users can now seamlessly evaluate their models with the MiniMax backend using skillopt-eval.

[Tanmay9223]

@microsoft-github-policy-service agree

[Yif-Yang]

Thanks for this @Tanmay9223! 🙏

Heads up on coordination: the eval-only backend wiring just landed via #85 (qwen + minimax configure_* calls in scripts/eval_only.py), and the MiniMax half of this PR overlaps almost entirely with what's now on main.

Comparing the two, #87 is effectively a superset of this PR — it carries the same eval_only.py minimax change plus the same json_utils.py change plus the new sleep verifier test. To avoid landing the same diff twice and creating a three-way conflict in eval_only.py, I'd suggest we close this one in favor of #87 and consolidate the review there.

Could you confirm you're OK closing #86? If there's anything in here that isn't also in #87, let me know and we'll make sure it's preserved. Really appreciate the backend coverage work.