Add MiniMax-M3 and MiniMax-M2.7 to benchmark model registry

[octo-patch]

Reason: The scheduled OpenRouter benchmark registry lists only the obsolete MiniMax M2.1 and M2.5 models and is missing the current MiniMax-M3 and MiniMax-M2.7 targets. The custom endpoint path also assigned every model the same 200,000-token, text-only metadata.

This PR adds openrouter/minimax/minimax-m3 and openrouter/minimax/minimax-m2.7 to .github/benchmark-models.yml. The existing M2.1 and M2.5 entries remain so historical leaderboard runs stay reproducible.

It also updates custom OpenAI-compatible endpoint configuration for the official global and China base URLs. MiniMax-M3 and MiniMax-M2.7 now receive their model-specific context windows, input modalities, reasoning support, and pricing; unrelated custom endpoints keep the existing generic defaults.

Checks:

• python3 tests/test_lib_agent_models.py (4 tests)
• python3 tests/test_lib_agent_judge.py (3 tests)
• ruff==0.15.6 on the touched and related test files
• Parsed .github/benchmark-models.yml with PyYAML and verified 35 unique model entries
• git diff --check

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (3 files)

• .github/benchmark-models.yml
• scripts/lib_agent.py
• tests/test_lib_agent_models.py

Previous Review Summary (commit 90a335f)

Current summary above is authoritative. Previous snapshots are kept for context only.

Previous review (commit 90a335f)

Status: No Issues Found | Recommendation: Merge

Files Reviewed (1 file)

• .github/benchmark-models.yml

---

_{Reviewed by claude-sonnet-5 · Input: 174 · Output: 4.3K · Cached: 248.6K}

[octo-patch]

Updated the custom endpoint configuration for MiniMax-M3 and MiniMax-M2.7 to use the official global and China endpoint metadata, including context windows, input modalities, reasoning support, and pricing, and pushed commit 3ece9ac. I ran 7 focused and existing unit tests, Ruff 0.15.6, registry YAML validation, and git diff --check.