chore: turn OpenAIMixin into a pydantic.BaseModel

- implement get_api_key instead of relying on LiteLLMOpenAIMixin.get_api_key
 - remove use of LiteLLMOpenAIMixin
 - add default initialize/shutdown methods to OpenAIMixin
 - remove __init__s to allow proper pydantic construction
 - remove dead code from vllm adapter and associated / duplicate unit tests
 - update vllm adapter to use openaimixin for model registration
 - remove ModelRegistryHelper from fireworks & together adapters
 - remove Inference from nvidia adapter
 - complete type hints on embedding_model_metadata
 - allow extra fields on OpenAIMixin, for model_store, __provider_id__, etc
 - new recordings for ollama

llama_stack/providers/remote/inference/cerebras/cerebras.py

@@ -11,7 +11,6 @@ from cerebras.cloud.sdk import AsyncCerebras
 from llama_stack.apis.inference import (
     ChatCompletionRequest,
     CompletionRequest,
-    Inference,
     OpenAIEmbeddingsResponse,
     TopKSamplingStrategy,
 )
@@ -27,14 +26,12 @@ from llama_stack.providers.utils.inference.prompt_adapter import (
 from .config import CerebrasImplConfig
 
 
-class CerebrasInferenceAdapter(
-    OpenAIMixin,
-    Inference,
-):
-    def __init__(self, config: CerebrasImplConfig) -> None:
-        self.config = config
+class CerebrasInferenceAdapter(OpenAIMixin):
+    config: CerebrasImplConfig
 
-        # TODO: make this use provider data, etc. like other providers
+    _cerebras_client: AsyncCerebras | None = None
+
+    async def initialize(self) -> None:
         self._cerebras_client = AsyncCerebras(
             base_url=self.config.base_url,
             api_key=self.config.api_key.get_secret_value(),
@@ -46,12 +43,6 @@ class CerebrasInferenceAdapter(
     def get_base_url(self) -> str:
         return urljoin(self.config.base_url, "v1")
 
-    async def initialize(self) -> None:
-        return
-
-    async def shutdown(self) -> None:
-        pass
-
     async def _get_params(self, request: ChatCompletionRequest | CompletionRequest) -> dict:
         if request.sampling_params and isinstance(request.sampling_params.strategy, TopKSamplingStrategy):
             raise ValueError("`top_k` not supported by Cerebras")