Fix CodeRabbit review issues: credential casing, client cleanup, and TypeScript interfaces

- Fix credential key casing mismatch in credential_service.py (LLM_PROVIDER vs llm_provider)
- Add proper OpenAI client cleanup to prevent resource leaks in llm_provider_service.py
- Add missing LLM_INSTANCE_NAME and OLLAMA_EMBEDDING_INSTANCE_NAME fields to RAGSettings interface

These fixes address critical CodeRabbit review comments and resolve TypeScript compilation errors.

python/src/server/services/credential_service.py

@@ -472,7 +472,7 @@ class CredentialService:
         try:
             # For now, we'll update the RAG strategy settings
             return await self.set_credential(
-                "llm_provider",
+                "LLM_PROVIDER",
                 provider,
                 category="rag_strategy",
                 description=f"Active {service_type} provider",

python/src/server/services/llm_provider_service.py

@@ -143,7 +143,8 @@ async def get_llm_client(provider: str | None = None, use_embedding_provider: bo
         raise
     finally:
         # Cleanup if needed
-        pass
+        if client and hasattr(client, "close"):
+            await client.close()
 
 
 async def _get_optimal_ollama_instance(instance_type: str | None = None,