feat: add Cheaper Inference as a provider option

[aiapienthusiast]

This PR adds Cheaper Inference as a provider option in the quickstart example.
It uses the same pattern as #270 (Eden AI).

Cheaper Inference is an OpenAI-compatible LLM gateway.
One API key gives access to models from several labs.

Usage:

export CHEAPER_INFERENCE_API_KEY=...
python atomic-examples/quickstart/quickstart/4_basic_chatbot_different_providers.py
# choose 9 or "cheaperinference"

Get a key at https://cheaperinference.com/signup.

Details

Files:

• atomic-examples/quickstart/quickstart/4_basic_chatbot_different_providers.py: add option 9, cheaperinference.
  It uses instructor.from_openai with base URL https://api.cheaperinference.com/v1 and model gpt-5.4-mini.
• README.md: add Cheaper Inference to the list of OpenAI-compatible APIs.

Tests:

• black --check (24.10) on the changed example: unchanged.
• flake8 on the changed example: clean.

Live check: gpt-5.4-mini returned a reply through setup_client("cheaperinference") and AtomicAgent.run().

[Eigenwise]

The missing-key path needs fixing before merge. The new branch passes None to OpenAI when CHEAPER_INFERENCE_API_KEY is absent. The locked SDK then falls back to OPENAI_API_KEY, so choosing this gateway can send an unrelated provider key to the third-party URL. Validate a nonempty vendor key before constructing the client, and test that missing-key path with a dummy unrelated OpenAI key and no network requests.

There are no tests of the actual new setup branch. Please add properly mocked selector, credential, request-shape and structured-response tests. The modified setup function also exceeds our per-function complexity bar; the existing core coverage job doesn't cover this example.

Please explain the distinct developer benefit of another named gateway entry beyond the existing OpenAI-compatible examples. A provider-neutral configuration example may cover the use case with less maintenance. I haven't verified this vendor's model/mode/token-parameter compatibility, so I'm not treating that as a confirmed failure.

[Eigenwise]

Source-only recheck of 2f28aa63: the missing/empty vendor-key path now rejects the key before constructing OpenAI, and setup_client is down to complexity 2. The added tests contain relevant selector, credential and request assertions. Those repairs look right in the source.

One measured production blocker remains: atomic-examples/quickstart/quickstart/4_basic_chatbot_different_providers.py:134-205, the new main, has native Radon 6.0.1 complexity 7. Our per-function CRAP requirement is strictly below 6, so this fails even with full coverage. Please simplify the interactive flow at meaningful boundaries. Keep the useful test assertions.

I haven't run this head's tests, measured its coverage or verified SDK/provider behavior. The links remain neutral; the earlier question about the distinct benefit of another named gateway entry is still open.

[Eigenwise]

I'm closing this under the policy in #273 and #286. We're not adding more named gateways to the quickstart; OpenAI-compatible endpoints are already covered by the generic setup. I should've caught that before asking for revisions. No further changes needed.