Populate resolved_model in the default OpenAI plugin from the response `model` field

Following #1117: set_resolved_model() shipped in 0.27, and plugins like llm-gemini (modelVersion) and llm-llama-server (/models) now record the resolved model. The built-in OpenAI plugin (llm/default_plugins/openai_models.py) doesn't call it, even though the OpenAI-compatible response already carries a model field that resolves aliases to the dated snapshot — exactly the "you asked for a *-latest alias, the API tells you which version actually answered" case from #1117.

Proposal: in the default OpenAI Chat/AsyncChat plugin, call response.set_resolved_model(...) from the response's own model field when it is present and differs from the requested model id. Zero extra calls (the field is already in every response body), and it makes resolved_model populated by default for the most common provider — surfacing silent alias→version resolution, and any gateway/relay substitution, directly in llm logs.

Context for the rationale (reading the served model on every turn as a zero-cost provenance signal): Engine Provenance, doi.org/10.5281/zenodo.22722524 — but the concrete ask here is just wiring the existing set_resolved_model method into the default OpenAI plugin.