feat: add MiniMax as LLM provider for evaluation and examples

[octo-patch]

Summary

Adds MiniMax as an LLM provider for the evaluation framework and a new integration example.

MiniMax offers OpenAI-compatible endpoints (https://api.minimax.io/v1) with 204K-context models (M2.7, M2.7-highspeed, M2.5, M2.5-highspeed), making it a drop-in alternative for any workflow that already uses the openai SDK.

Changes

Area	What changed
evaluation/agent_loop.py	New OpenAIAgentLoop — works with standard OpenAI and any compatible API. Includes MiniMax-specific temperature clamping (> 0) and <think> tag stripping for M2.x models.
evaluation/main.py	--agent openai, --openai-base-url, --openai-model CLI flags so evaluations can run against MiniMax (or any OpenAI-compat provider) without code changes.
examples/minimax-integration/	Standalone example showing MiniMax function-calling with sandbox code execution (mirrors openai-integration).
evaluation/tests/	19 unit tests + 4 live integration tests (skipped when MINIMAX_API_KEY is not set).
READMEs	Updated README.md, examples/README.md, and evaluation/README.md with MiniMax usage docs.

Usage

# Run evaluation with MiniMax
export OPENAI_API_KEY="your_minimax_key"
uv run main.py --agent openai \
    --openai-base-url https://api.minimax.io/v1 \
    --openai-model MiniMax-M2.7

# Run the example
cd examples/minimax-integration
export MINIMAX_API_KEY="your_key"
uv run main.py

Test plan

• 19 unit tests pass (python -m pytest evaluation/tests/test_openai_agent_loop.py)
• 4 integration tests pass with live MiniMax API key
• Existing AzureOpenAIAgentLoop behaviour unchanged (verified by unit test)
• Run examples/minimax-integration/main.py against a live sandbox

13 files changed, 1019 additions(+), 12 deletions(-)