Optimizing Large Language Models for OpenAPI Code Completion