overriding model name at runtime using model settings

[mattbrandman]

Idea for enabling model name override when making temporal calls within a provider

Partially addresses #3407

[mattbrandman]

@DouweM this would be for working within a provider would require updates to any provider this should be allowed on. Might take a look at how to enable multiple providers as well like discussed

[DouweM]

@mattbrandman A good amount of request payload building logic in OpenAIChatModel branches based on self.profile, which depends on the model (name), so it's not sufficient to just change the model name we sent to the API without also changing self.profile and the payload as appropriate.

We could fix that, add a model_name arg to the base ModelSettings, and support it in every model class, but I don't really want to break the expectation that a model class instance represents a specific named model, "just" for Temporal.

Definitely open to other ideas in that issue, but I think the one I sketched is the most in line with Pydantic AI expectations.