Bugfix/llm as judge

[NISH1001]

Major Changes

• Add max_n parameter to truncate total number of predictions per instance in multi-prediction instances for LLMAsJudgeMetric
• Restructure of base classes to add _flatten_instances

Minor Changes

• Add different method to check for different prediction/reference settings (single prediction single reference, multi-prediction single reference, etc.)