feat(eval): add optional Cheaper Inference reader

[aiapienthusiast]

This PR adds an optional --cheaperinference reader to eval/run_bench.py.
It uses the same pattern as #159.

Cheaper Inference is an OpenAI-compatible LLM gateway.
gpt-5.4 and gpt-5.4-mini accept image input.

Cheaper Inference is one of the fastest-growing AI routers.
Each model costs 15–60% less than the list price of its lab.

Details

Behavior:

• Uses the existing OpenAI-compatible client.
• Reads the key only from --api-key or CHEAPER_INFERENCE_API_KEY.
• Sends model IDs unchanged.
• Stops with an error when combined with another provider flag, --litellm or --api-base.

Files:

• eval/lib/model_config.py
• eval/run_bench.py
• eval/README.md
• tests/test_cheaperinference_client.py (new)

Tests:

• ruff check . and ruff format --check .: passed.
• New tests and model config tests: passed.

[vercel]

@aiapienthusiast is attempting to deploy a commit to the andylizf's projects Team on Vercel.

A member of the Team first needs to authorize it.