Google Gemini/Gemma models sometimes append "<|file_separator|>" at the end of autocomplete completions, forcing users to remove it manually. Update postprocessCompletion to detect when llm.model includes "gemini" or "gemma" and the completion ends with "<|file_separator|>", then slice off the trailing 18 characters. See #6067