Examining the Robustness of Evaluation Metrics for Patent Retrieval with Incomplete Relevance Judgements