feat: use SecretStr for inference provider auth credentials

- RemoteInferenceProviderConfig now has auth_credential: SecretStr
- the default alias is api_key (most common name)
- some providers override to use api_token (RunPod, vLLM, Databricks)
- some providers exclude it (Ollama, TGI, Vertex AI)

llama_stack/providers/remote/inference/gemini/config.py

@@ -21,11 +21,6 @@ class GeminiProviderDataValidator(BaseModel):
 
 @json_schema_type
 class GeminiConfig(RemoteInferenceProviderConfig):
-    api_key: str | None = Field(
-        default=None,
-        description="API key for Gemini models",
-    )
-
     @classmethod
     def sample_run_config(cls, api_key: str = "${env.GEMINI_API_KEY:=}", **kwargs) -> dict[str, Any]:
         return {

llama_stack/providers/remote/inference/gemini/gemini.py

@@ -17,8 +17,5 @@ class GeminiInferenceAdapter(OpenAIMixin):
         "text-embedding-004": {"embedding_dimension": 768, "context_length": 2048},
     }
 
-    def get_api_key(self) -> str:
-        return self.config.api_key or ""
-
     def get_base_url(self):
         return "https://generativelanguage.googleapis.com/v1beta/openai/"