Add yolo-auto/yolo and yolo-auto/yolo-small model metadata and settings

[harryvgiunta]

This PR adds model metadata and settings for the two stable public aliases of Yolo-Auto, a flat-rate OpenAI-compatible inference router: yolo-auto/yolo (vision + reasoning) and yolo-auto/yolo-small (text-only). Supersedes #5747, which used openai/ keys; I re-keyed them to the real provider prefix.

This PR is blocked on BerriAI/litellm#42545 (registering yolo-auto in litellm, which is exactly the pattern this PR needs for routing). The yolo-auto/ prefix only routes once that registration ships in a litellm release. Until then this PR is harmless-but-inert (entries only add --list-models matches and metadata); marking it draft on purpose.

• I have followed the instructions in the CONTRIBUTING document
• I have run all tests and they work as expected (tests/basic/test_models.py + tests/basic/test_model_info_manager.py: 27 passed)
• My code passes the pre-commit status checks (pre-commit run --all-files)
• I have added tests to cover the new functionality (metadata/settings are data files; the manager tests cover the loading path)

If this change required a real-world end-to-end manual test using a real LLM API, here are the details explaining how you did this and what you found:

With a litellm checkout containing BerriAI/litellm#42545 and YOLO_AUTO_API_KEY set:

1. aider --model yolo-auto/yolo resolves: context 262144, vision true, weak/editor model yolo-auto/yolo-small
2. --list-models | grep yolo lists yolo-auto/yolo and yolo-auto/yolo-small
3. A real chat completion round-trips through https://yolo-auto.com/v1/chat/completions (model echoes content, finish_reason: stop, real usage counts)

Without the litellm registration (released litellm today), step 1 warns and step 3 fails with LLM Provider NOT provided, which is why this is a draft.

Summary

• aider/resources/model-metadata.json: exact-key rows for yolo-auto/yolo and yolo-auto/yolo-small (262144 context, 32768 output cap, zero per-token cost; the service is subscription flat-rate). litellm_provider: yolo-auto, so both --list-models fuzzy matching and get_model_from_cached_json_db find them.
• aider/resources/model-settings.yml: diff edit format, repo map, examples-as-system-message, with yolo-auto/yolo-small as the weak + editor model for both entries (same pairing style as the other diff-format entries in the file).

Relevant issues

None; new provider models. Related: BerriAI/litellm#42545